{
  "proof": {
    "phase": "sealed",
    "ranking": [
      {
        "genome": {
          "id": "sealed-selected",
          "general": "Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.",
          "inline": [
            "Use the quietest intelligible close whisper. Carry the full intended delivery through the global direction.",
            "Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Carry the full intended delivery through the global direction.",
            "Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction."
          ],
          "cfg": 2,
          "recipe": "P08",
          "timing_syntax": "canonical",
          "canonical_labels": true,
          "neutral_event": "keep",
          "parent": "cfg-2"
        },
        "takes": 100,
        "fitness": 0.6624451107135754,
        "steering_v2": 0.6624451107135754,
        "ce_mean": 5.835107021331787,
        "genuineness_mean": 4.345739071369171,
        "mean_wer": 0.015857734542106067,
        "valid_fraction": 1,
        "identity_delta_vs_fixed_g1": -0.04899539925158024,
        "passed_speech_identity_gates": true,
        "passed_development_gates": true,
        "family_hit_all_fraction": null,
        "nominal_hit_all_fraction": null,
        "status": "SEALED_TEST"
      },
      {
        "genome": {
          "id": "sealed-g1",
          "general": "Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.",
          "inline": [
            "Use the quietest intelligible close whisper. Carry the full intended delivery through the global direction.",
            "Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Carry the full intended delivery through the global direction.",
            "Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction."
          ],
          "cfg": 1.0,
          "recipe": "P08",
          "timing_syntax": "canonical",
          "canonical_labels": true,
          "neutral_event": "keep",
          "parent": "cfg-2"
        },
        "takes": 100,
        "fitness": 0.5083850864304328,
        "steering_v2": 0.5083850864304328,
        "ce_mean": 6.345443511009217,
        "genuineness_mean": 4.52385311126709,
        "mean_wer": 0.011474699956554795,
        "valid_fraction": 1,
        "identity_delta_vs_fixed_g1": 0.0,
        "passed_speech_identity_gates": true,
        "passed_development_gates": true,
        "family_hit_all_fraction": null,
        "nominal_hit_all_fraction": null,
        "status": "SEALED_TEST"
      }
    ],
    "takes": 200,
    "reward_version": "ears-steering-v2",
    "updated": 1793050410.9406598
  },
  "records": [
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "e0fbdd64407efa40453d370ae06374d4a4b383309db7f9a4da25ededb74f5a4a",
      "id": "dc396e5f30c0a426f04ffd466af9",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 6.38003396987915,
      "genuineness": 4.444586753845215,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.05814938619732857,
      "fitness": 0.4822092976539126,
      "legacy_fitness": 0.4822092976539126,
      "target_cdf": 0.2750346673267665,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.8114793300628662,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4454925877909376,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.8114793300628662,
          "normalized": 0.2750346673267665,
          "contribution": 0.19252426712873655
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8091652393341064,
          "normalized": 0.8432277355406702,
          "contribution": 0.25296832066220104
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/dc/dc396e5f30c0a426f04ffd466af9.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "31a580feee3f3d2b18bde2e0440a28e93d883eee0dd175986d27066364a98ee8",
      "id": "80aaa1209ed552a7dfa85fe0a1fe",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 6.3641486167907715,
      "genuineness": 4.482243061065674,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.011902354657649994,
      "fitness": 0.5560048868875423,
      "legacy_fitness": 0.5560048868875423,
      "target_cdf": 0.42028208784674,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.4453237056732178,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5499238707457959,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.4453237056732178,
          "normalized": 0.42028208784674,
          "contribution": 0.29419746149271797
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7752411365509033,
          "normalized": 0.8524213641769265,
          "contribution": 0.25572640925307794
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/80/80aaa1209ed552a7dfa85fe0a1fe.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "3fb61e8c3df23ca8ec88259177c84ecefe38fdb02fc9a4c3a51fa068fffdc56c",
      "id": "e609f0076be42f8fee5fcff3fcee",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 6.268836498260498,
      "genuineness": 4.432034492492676,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.029951337724924088,
      "fitness": 0.5584251847831478,
      "legacy_fitness": 0.5584251847831478,
      "target_cdf": 0.4340723369455479,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.5015267133712769,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.560115397305766,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.5015267133712769,
          "normalized": 0.4340723369455479,
          "contribution": 0.30385063586188354
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.768456220626831,
          "normalized": 0.8542158714796083,
          "contribution": 0.2562647614438825
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/e6/e609f0076be42f8fee5fcff3fcee.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "0ee5e1532ca1fc650fea69912aecc2c834c76af27ff597100b47c7dc084f7e05",
      "id": "5981ce2afe91f276e46fb1a5e9f5",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 6.427689552307129,
      "genuineness": 4.532451152801514,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.1059541255235672,
      "fitness": 0.5835881571263976,
      "legacy_fitness": 0.5835881571263976,
      "target_cdf": 0.4680875739039791,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.6389117240905762,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5951018216235312,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.6389117240905762,
          "normalized": 0.4680875739039791,
          "contribution": 0.32766130173278535
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6124054193496704,
          "normalized": 0.8914683996358195,
          "contribution": 0.26744051989074585
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/59/5981ce2afe91f276e46fb1a5e9f5.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "c35a3bce3438369174db0bd229806b2e6e48ea4df7d33c4963fa055b8d9679ed",
      "id": "d982785c1e5a671474bac14712c5",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 5.967014789581299,
      "genuineness": 4.2249250411987305,
      "wer": 0.06451612903225806,
      "identity": null,
      "voiceclap": 0.058186642825603485,
      "fitness": 0.6078299104850727,
      "legacy_fitness": 0.6078299104850727,
      "target_cdf": 0.649070603067288,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.3789172172546387,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.67608445510922,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.3789172172546387,
          "normalized": 0.649070603067288,
          "contribution": 0.4543494221471016
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.1280517578125,
          "normalized": 0.7391167765403948,
          "contribution": 0.2217350329621184
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d9/d982785c1e5a671474bac14712c5.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "d90538a9d7f3cb648d7cb50b6b08ada81c86e0c91101dd4d60e34f398306e6fe",
      "id": "ceca1cdbe4d081f56e2aa7cfec4e",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 6.109982967376709,
      "genuineness": 4.344170093536377,
      "wer": 0.0,
      "identity": 0.720887303352356,
      "voiceclap": 0.09753774851560593,
      "fitness": 0.5950487968241218,
      "legacy_fitness": 0.5950487968241218,
      "target_cdf": 0.5225842708180436,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.8574787378311157,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6215353988257084,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.8574787378311157,
          "normalized": 0.5225842708180436,
          "contribution": 0.36580898957263047
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7752411365509033,
          "normalized": 0.8524213641769265,
          "contribution": 0.25572640925307794
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/ce/ceca1cdbe4d081f56e2aa7cfec4e.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "ad667b08ed46d49f2fbf8828ccee20769fbb54a7ab8c6c4932445804cba19895",
      "id": "8ce44f0b77323a8441a515e540d4",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 6.268836498260498,
      "genuineness": 4.7081804275512695,
      "wer": 0.0,
      "identity": 0.6812404990196228,
      "voiceclap": 0.12300090491771698,
      "fitness": 0.6741609852580681,
      "legacy_fitness": 0.6741609852580681,
      "target_cdf": 0.6425317766405054,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.350815773010254,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7303301153638737,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.350815773010254,
          "normalized": 0.6425317766405054,
          "contribution": 0.4497722436483538
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.3681519031524658,
          "normalized": 0.9351929057184,
          "contribution": 0.28055787171552
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/8c/8ce44f0b77323a8441a515e540d4.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "c7b84b84343c55a9ef785cd3c5be89d7252083140ce9d635bff69b73ba4bed55",
      "id": "1c69ab42af1cd20b95dba1c603ae",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 6.09409761428833,
      "genuineness": 4.607763767242432,
      "wer": 0.0,
      "identity": 0.6822501420974731,
      "voiceclap": 0.04525743052363396,
      "fitness": 0.6470865793017381,
      "legacy_fitness": 0.6470865793017381,
      "target_cdf": 0.6054879639521904,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.1946964263916016,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6700462900078777,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.1946964263916016,
          "normalized": 0.6054879639521904,
          "contribution": 0.4238415747665333
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.887190580368042,
          "normalized": 0.820682384137815,
          "contribution": 0.2462047152413445
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/1c/1c69ab42af1cd20b95dba1c603ae.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "5dc8d5b7d8875d671274ab5b08062dbc73ba5c9be2cadf8b856b3f5ddbaff7d6",
      "id": "a47da3d5d326f00ffe59a792e931",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 6.260893821716309,
      "genuineness": 4.607763767242432,
      "wer": 0.0,
      "identity": 0.5826681852340698,
      "voiceclap": 0.12523408234119415,
      "fitness": 0.6723231446159931,
      "legacy_fitness": 0.6723231446159931,
      "target_cdf": 0.6476212842093014,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.3726725578308105,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7411135530776343,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.3726725578308105,
          "normalized": 0.6476212842093014,
          "contribution": 0.45333489894651097
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.1713922023773193,
          "normalized": 0.9592621804370778,
          "contribution": 0.2877786541311233
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/a4/a47da3d5d326f00ffe59a792e931.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "726431e89ce1713898fc2d0cb371ee0ae527e424cf49e9c87b8386e2002e3b56",
      "id": "c2f706cd066093e882e47a678a31",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 5.990842819213867,
      "genuineness": 4.3881025314331055,
      "wer": 0.0,
      "identity": 0.6779211759567261,
      "voiceclap": 0.1064876839518547,
      "fitness": 0.5732618950844608,
      "legacy_fitness": 0.5732618950844608,
      "target_cdf": 0.4813064382554696,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.6919922828674316,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6016599390518089,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.6919922828674316,
          "normalized": 0.4813064382554696,
          "contribution": 0.33691450677882867
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6531143188476562,
          "normalized": 0.8824847742432673,
          "contribution": 0.2647454322729802
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/c2/c2f706cd066093e882e47a678a31.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "d98b8c7933f81d303477bc9c29bcf78fee9777634db9feeffe89e2e25bf2062b",
      "id": "c800fd5a81f235407f4824bfe786",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 6.443574905395508,
      "genuineness": 4.419482707977295,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.06989045441150665,
      "fitness": 0.60497042037125,
      "legacy_fitness": 0.60497042037125,
      "target_cdf": 0.51947186980795,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.8449891805648804,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5731217640478086,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.8449891805648804,
          "normalized": 0.51947186980795,
          "contribution": 0.36363030886556497
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.2332162857055664,
          "normalized": 0.6983048506074785,
          "contribution": 0.20949145518224355
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/c8/c800fd5a81f235407f4824bfe786.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "f43b25b1e704bd0779364a7d0322afd1bdacfec487fc664afdd06bffbb644dbd",
      "id": "4f7e1ddafe3f20718258855291ed",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 5.792276382446289,
      "genuineness": 4.080576419830322,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.03848133981227875,
      "fitness": 0.6088801475139216,
      "legacy_fitness": 0.6088801475139216,
      "target_cdf": 0.5880984409196686,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.1228814125061035,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6525003644427507,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.1228814125061035,
          "normalized": 0.5880984409196686,
          "contribution": 0.411668908643768
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.9448615312576294,
          "normalized": 0.8027715193299423,
          "contribution": 0.24083145579898269
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/4f/4f7e1ddafe3f20718258855291ed.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "4bcad42e61229de75a1aa8a5541e662ff469b627a1ec4125b19b38dea3c3d7f2",
      "id": "7cb398823fb52b99920a4770aff6",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 5.95112943649292,
      "genuineness": 4.049196243286133,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.057918231934309006,
      "fitness": 0.606207661340322,
      "legacy_fitness": 0.606207661340322,
      "target_cdf": 0.5774258305821535,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.0791680812835693,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.630432924266786,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.0791680812835693,
          "normalized": 0.5774258305821535,
          "contribution": 0.4041980814075074
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0873427391052246,
          "normalized": 0.7541161428642621,
          "contribution": 0.2262348428592786
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/7c/7cb398823fb52b99920a4770aff6.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "a840b4b01682217fc047cdca15f9caaa6ba479fe75f9e2e3ab2e01a26467957d",
      "id": "744b92c5bbfd1168036da936b5fa",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 5.760505676269531,
      "genuineness": 4.105680465698242,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.007902730256319046,
      "fitness": 0.6689665096575333,
      "legacy_fitness": 0.6689665096575333,
      "target_cdf": 0.7077676966934032,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.641197681427002,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6991308929585595,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.641197681427002,
          "normalized": 0.7077676966934032,
          "contribution": 0.4954373876853822
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.280709981918335,
          "normalized": 0.6789783509105908,
          "contribution": 0.20369350527317723
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/74/744b92c5bbfd1168036da936b5fa.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "a91589d86515de7f6a6964b06b4c3298e0a3bc21dea914e3c85c4f5f546069f3",
      "id": "4eb84b20046b8f3ed5ea668c6c73",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 6.0146708488464355,
      "genuineness": 4.18099308013916,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.016684111207723618,
      "fitness": 0.6190040295579753,
      "legacy_fitness": 0.6190040295579753,
      "target_cdf": 0.5888584266620323,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.1260039806365967,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6420767966107916,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.1260039806365967,
          "normalized": 0.5888584266620323,
          "contribution": 0.4122008986634226
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0534186363220215,
          "normalized": 0.7662529931578967,
          "contribution": 0.229875897947369
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/4e/4eb84b20046b8f3ed5ea668c6c73.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "f0bf9d9e62b8b68f4d1a7e5afa8bcd6eb9f34f95b34364cc63c6d56b24b8781d",
      "id": "35f3a54f850e2298dd6a7dbc9f69",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 6.022613525390625,
      "genuineness": 4.030367851257324,
      "wer": 0.0,
      "identity": 0.5907247066497803,
      "voiceclap": 0.09286336600780487,
      "fitness": 0.5782332668411292,
      "legacy_fitness": 0.5782332668411292,
      "target_cdf": 0.51947186980795,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.8449891805648804,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5977407933086534,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.8449891805648804,
          "normalized": 0.51947186980795,
          "contribution": 0.36363030886556497
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.012709617614746,
          "normalized": 0.7803682814769612,
          "contribution": 0.23411048444308835
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/35/35f3a54f850e2298dd6a7dbc9f69.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "49305371c31fcc93f5ee684e59383215d0c15d4d84e9006085c8ab0e20a109cf",
      "id": "b9cc00ae35f342e5f16c3a843c32",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 5.95112943649292,
      "genuineness": 4.256305694580078,
      "wer": 0.0,
      "identity": 0.5725533366203308,
      "voiceclap": 0.05330374091863632,
      "fitness": 0.6148372218109031,
      "legacy_fitness": 0.6148372218109031,
      "target_cdf": 0.5774258305821535,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.0791680812835693,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6460006562235984,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.0791680812835693,
          "normalized": 0.5774258305821535,
          "contribution": 0.4041980814075074
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.9346842765808105,
          "normalized": 0.8060085827203032,
          "contribution": 0.24180257481609094
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/b9/b9cc00ae35f342e5f16c3a843c32.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "536e829f003c6d18e9a2d50852c55777227c3f40baea13ece384656777a8c08d",
      "id": "cc110a3887940ab1d0e67194a81b",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 5.959072113037109,
      "genuineness": 4.074300289154053,
      "wer": 0.0,
      "identity": 0.7059857845306396,
      "voiceclap": 0.029686085879802704,
      "fitness": 0.600165137707588,
      "legacy_fitness": 0.600165137707588,
      "target_cdf": 0.5628516456671493,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.0198428630828857,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6301675985496518,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.0198428630828857,
          "normalized": 0.5628516456671493,
          "contribution": 0.3939961519670045
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.9923553466796875,
          "normalized": 0.7872381552754909,
          "contribution": 0.23617144658264727
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/cc/cc110a3887940ab1d0e67194a81b.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "296b23db9f575e058e8044b99deef5fa7affb1636f964cfb01817b9eb6b95161",
      "id": "a40f3e0679deacdbe2011f29c65d",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 2,
      "generation": null,
      "ce": 5.895530700683594,
      "genuineness": 4.306513786315918,
      "wer": 0.0,
      "identity": 0.4921066164970398,
      "voiceclap": 0.06634975224733353,
      "fitness": 0.6571192633700965,
      "legacy_fitness": 0.6571192633700965,
      "target_cdf": 0.6605858428463536,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.42887544631958,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7035663285041078,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.42887544631958,
          "normalized": 0.6605858428463536,
          "contribution": 0.4624100899924475
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.9414691925048828,
          "normalized": 0.803854128372201,
          "contribution": 0.2411562385116603
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/a4/a40f3e0679deacdbe2011f29c65d.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "27cba5a3f70ae42599d545ecf6d935422312a8b8972a36ed8e720f8faef1928e",
      "id": "e32773f654877356583166cbc579",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.84,
      "cfg": 2,
      "generation": null,
      "ce": 5.637393951416016,
      "genuineness": 4.281409740447998,
      "wer": 0.0,
      "identity": 0.539197564125061,
      "voiceclap": 0.06578640639781952,
      "fitness": 0.6122352640706592,
      "legacy_fitness": 0.6122352640706592,
      "target_cdf": 0.5858166855331843,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.1135144233703613,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6679249895342716,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.1135144233703613,
          "normalized": 0.5858166855331843,
          "contribution": 0.410071679873229
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7481017112731934,
          "normalized": 0.8595110322034752,
          "contribution": 0.2578533096610426
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/e3/e32773f654877356583166cbc579.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "508fcc4c37613a9b3f50e9f7bf5f968e298888bcd1d4a3e6b44fec276b9bed75",
      "id": "d0ff13e03a99c9db868e62e5ba64",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 2,
      "generation": null,
      "ce": 6.316492557525635,
      "genuineness": 4.243753433227539,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.02247856929898262,
      "fitness": 0.6635956476659977,
      "legacy_fitness": 0.6635956476659977,
      "target_cdf": 0.6577205480200855,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.4163858890533447,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6594369461054775,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.4163858890533447,
          "normalized": 0.6577205480200855,
          "contribution": 0.46040438361405983
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.318026542663574,
          "normalized": 0.6634418749713924,
          "contribution": 0.1990325624914177
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d0/d0ff13e03a99c9db868e62e5ba64.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "c95c6e80231999b991a806f5cefb80af5afcfcaf6d3d90107bb5e56571e4602c",
      "id": "c355ecd2010866da0fd59cc29319",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 2,
      "generation": null,
      "ce": 5.831989288330078,
      "genuineness": 3.9801597595214844,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.04871414229273796,
      "fitness": 0.5899838837250364,
      "legacy_fitness": 0.5899838837250364,
      "target_cdf": 0.5566883230734452,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.994863748550415,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.61141685911353,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.994863748550415,
          "normalized": 0.5566883230734452,
          "contribution": 0.3896818261514116
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.1280517578125,
          "normalized": 0.7391167765403948,
          "contribution": 0.2217350329621184
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/c3/c355ecd2010866da0fd59cc29319.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "23ff5c0fbbfc96e905737c32ee86f243be3cfc59c5c67723ad52088e0df79a94",
      "id": "e17aad720bd50899c36bc1995747",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 2,
      "generation": null,
      "ce": 6.157639026641846,
      "genuineness": 4.344170093536377,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.07755377143621445,
      "fitness": 0.6638083369067713,
      "legacy_fitness": 0.6638083369067713,
      "target_cdf": 0.6577205480200855,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.4163858890533447,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.712807812416763,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.4163858890533447,
          "normalized": 0.6577205480200855,
          "contribution": 0.46040438361405983
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8159499168395996,
          "normalized": 0.8413447626756774,
          "contribution": 0.2524034288027032
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/e1/e17aad720bd50899c36bc1995747.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "ce5b9d36103e577bf349dd87c1ba7a0cb159c6ad12a91faa8ea57ae3c7951cd3",
      "id": "479b84eef7d5ca7bbd1fd316d49f",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 2,
      "generation": null,
      "ce": 5.649308204650879,
      "genuineness": 4.21237325668335,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.044529110193252563,
      "fitness": 0.6624794815518675,
      "legacy_fitness": 0.6624794815518675,
      "target_cdf": 0.6914624481475787,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.56626033782959,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7267876215457841,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.56626033782959,
          "normalized": 0.6914624481475787,
          "contribution": 0.4840237137033051
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.9245071411132812,
          "normalized": 0.8092130261415968,
          "contribution": 0.24276390784247903
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/47/479b84eef7d5ca7bbd1fd316d49f.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "4f60a5131c3b8a1316b7ef68d878246097032b5494788bc97ed5b6ae903456f3",
      "id": "c77edef1161852a0e919e5c55f41",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 2,
      "generation": null,
      "ce": 5.839932441711426,
      "genuineness": 4.268857479095459,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.06633234769105911,
      "fitness": 0.6344597288997486,
      "legacy_fitness": 0.6344597288997486,
      "target_cdf": 0.6211847124559708,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.2602665424346924,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.684928810279879,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.2602665424346924,
          "normalized": 0.6211847124559708,
          "contribution": 0.4348292987191796
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.84308922290802,
          "normalized": 0.8336650385356648,
          "contribution": 0.25009951156069943
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/c7/c77edef1161852a0e919e5c55f41.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "013e2fb4c9b309cde777bb54b6265fcb166262b00a196cb20e273f1e83533341",
      "id": "6c652d8bc1895f6f4cdaeca8a818",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 2,
      "generation": null,
      "ce": 5.863759994506836,
      "genuineness": 4.15588903427124,
      "wer": 0.0,
      "identity": 0.6021683216094971,
      "voiceclap": 0.06801512837409973,
      "fitness": 0.6273695307137473,
      "legacy_fitness": 0.6273695307137473,
      "target_cdf": 0.6152269755128829,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.2352874279022217,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6654605578044336,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.2352874279022217,
          "normalized": 0.6152269755128829,
          "contribution": 0.430658882859018
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.005924940109253,
          "normalized": 0.7826722498180521,
          "contribution": 0.2348016749454156
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/6c/6c652d8bc1895f6f4cdaeca8a818.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "69c3d6ccefbba7687134a7b03bf75c4622bfebbcaa678a56cd725f2df1318c59",
      "id": "054a6e6d42c70e94db80d7249807",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 2,
      "generation": null,
      "ce": 5.816103935241699,
      "genuineness": 4.024092197418213,
      "wer": 0.0,
      "identity": 0.6741227507591248,
      "voiceclap": 0.02150392159819603,
      "fitness": 0.6025502069563298,
      "legacy_fitness": 0.6025502069563298,
      "target_cdf": 0.578954200699057,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.0854129791259766,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6012895997464212,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.0854129791259766,
          "normalized": 0.578954200699057,
          "contribution": 0.4052679404893399
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.341773509979248,
          "normalized": 0.6534055308569375,
          "contribution": 0.19602165925708126
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/05/054a6e6d42c70e94db80d7249807.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "ae2bcbf0e80b791aead4ea9fe22aee586a932b169a7b50fd7c6b9eaa80c291ae",
      "id": "f53f7d5137aa92800963d9305f3f",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 2,
      "generation": null,
      "ce": 6.1894097328186035,
      "genuineness": 4.331617832183838,
      "wer": 0.0,
      "identity": 0.31064939498901367,
      "voiceclap": 0.1038978174328804,
      "fitness": 0.7259803234202672,
      "legacy_fitness": 0.7259803234202672,
      "target_cdf": 0.7815220075176177,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 3.009639263153076,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.8011502492889093,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 3.009639263153076,
          "normalized": 0.7815220075176177,
          "contribution": 0.5470654052623324
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7955955266952515,
          "normalized": 0.84694948008859,
          "contribution": 0.254084844026577
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/f5/f53f7d5137aa92800963d9305f3f.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "bc9efe4ef239840d70d03f44fd4ee6b3ded2bdc582b608af7f34456546589af2",
      "id": "afb0442d6ab20f5b467be5138df5",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 2,
      "generation": null,
      "ce": 6.260893821716309,
      "genuineness": 4.331617832183838,
      "wer": 0.0,
      "identity": 0.5823388695716858,
      "voiceclap": 0.09898950159549713,
      "fitness": 0.6715692371561832,
      "legacy_fitness": 0.6715692371561832,
      "target_cdf": 0.6691256305445644,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.466344118118286,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7187793113185155,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.466344118118286,
          "normalized": 0.6691256305445644,
          "contribution": 0.46838794138119505
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8396968841552734,
          "normalized": 0.8346378997910682,
          "contribution": 0.2503913699373204
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/af/afb0442d6ab20f5b467be5138df5.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "0aba88c5ae398e3acb458d0635a78ea3780c2873fe2b284b44d0202c81dde8d4",
      "id": "9adaffda9dbaf9a27901bd9c75d0",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 2,
      "generation": null,
      "ce": 5.824047088623047,
      "genuineness": 4.356721878051758,
      "wer": 0.0,
      "identity": 0.5897144079208374,
      "voiceclap": 0.037485718727111816,
      "fitness": 0.6545543278832309,
      "legacy_fitness": 0.6545543278832309,
      "target_cdf": 0.6548461448309961,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.4038963317871094,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6955799781346347,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.4038963317871094,
          "normalized": 0.6548461448309961,
          "contribution": 0.45839230138169723
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.982177972793579,
          "normalized": 0.7906255891764582,
          "contribution": 0.23718767675293745
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/9a/9adaffda9dbaf9a27901bd9c75d0.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "838eea514b0e6728de99a5812f0ada141ddc15b93cd75ca91bbc0386f09eee57",
      "id": "3742c6840a6b30ac87118e1427c6",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 2,
      "generation": null,
      "ce": 5.522225379943848,
      "genuineness": 4.256305694580078,
      "wer": 0.12903225806451613,
      "identity": null,
      "voiceclap": 0.02690565213561058,
      "fitness": 0.6228749451262393,
      "legacy_fitness": 0.6228749451262393,
      "target_cdf": 0.7995020193369433,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 3.109555721282959,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7951400973156957,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 3.109555721282959,
          "normalized": 0.7995020193369433,
          "contribution": 0.5596514135358602
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.9991400241851807,
          "normalized": 0.7849622792661184,
          "contribution": 0.2354886837798355
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/37/3742c6840a6b30ac87118e1427c6.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "dcc69f28730d1fa45cf54117a55edf07fe334185029076d287023c3fb55f6c1e",
      "id": "86ae1d205a29a4d65864e92b7714",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 2,
      "generation": null,
      "ce": 5.343515396118164,
      "genuineness": 4.268857479095459,
      "wer": 0.06451612903225806,
      "identity": null,
      "voiceclap": 0.0873035192489624,
      "fitness": 0.6475196781350294,
      "legacy_fitness": 0.6475196781350294,
      "target_cdf": 0.7614385222593032,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.9034781455993652,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7842672894159293,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.9034781455993652,
          "normalized": 0.7614385222593032,
          "contribution": 0.5330069655815122
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8295196294784546,
          "normalized": 0.8375344127813906,
          "contribution": 0.25126032383441715
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/86/86ae1d205a29a4d65864e92b7714.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "2b8d67d1b9e2c88bb854d7b0392fd9df37037b54aabef5e09c2f34845d270747",
      "id": "31537c59b92b4b56f5f36142cad9",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 2,
      "generation": null,
      "ce": 5.756534099578857,
      "genuineness": 4.494794845581055,
      "wer": 0.06451612903225806,
      "identity": null,
      "voiceclap": 0.0917959064245224,
      "fitness": 0.6483152496027729,
      "legacy_fitness": 0.6483152496027729,
      "target_cdf": 0.7236603501435066,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.716135025024414,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7773611117762433,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.716135025024414,
          "normalized": 0.7236603501435066,
          "contribution": 0.5065622451004546
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.5581269264221191,
          "normalized": 0.9026628889192958,
          "contribution": 0.27079886667578873
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/31/31537c59b92b4b56f5f36142cad9.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "7128fd33d3392fce4bad553f6e5b1b155b3c10aa7ea31903e67e395981967fe5",
      "id": "668a59de1d4943724fa9c779f9e6",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 2,
      "generation": null,
      "ce": 5.665193557739258,
      "genuineness": 4.507347106933594,
      "wer": 0.06451612903225806,
      "identity": null,
      "voiceclap": 0.0675278753042221,
      "fitness": 0.6423561260851396,
      "legacy_fitness": 0.6423561260851396,
      "target_cdf": 0.7144411603494439,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.672421455383301,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7510805987217285,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.672421455383301,
          "normalized": 0.7144411603494439,
          "contribution": 0.5001088122446107
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8329119682312012,
          "normalized": 0.8365726215903928,
          "contribution": 0.2509717864771178
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/66/668a59de1d4943724fa9c779f9e6.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "0ac3df6009d8108c373c37b13721bebf5bca097a424c1a1766c167664747b671",
      "id": "17beeba6effc28b7eace8fc0cbe0",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 2,
      "generation": null,
      "ce": 5.736677646636963,
      "genuineness": 4.570107460021973,
      "wer": 0.03225806451612903,
      "identity": null,
      "voiceclap": 0.07178813219070435,
      "fitness": 0.632204981590591,
      "legacy_fitness": 0.632204981590591,
      "target_cdf": 0.6388807912868757,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.3352038860321045,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7142184946177561,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.3352038860321045,
          "normalized": 0.6388807912868757,
          "contribution": 0.44721655390081294
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6191902160644531,
          "normalized": 0.8900064690564774,
          "contribution": 0.2670019407169432
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/17/17beeba6effc28b7eace8fc0cbe0.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "1d11c01674aaf4d6220c196cc9577c1877f645eed3b453081217dc5703688566",
      "id": "46769a51d397a5a603d2a54d66da",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 2,
      "generation": null,
      "ce": 5.426913261413574,
      "genuineness": 4.406930446624756,
      "wer": 0.06451612903225806,
      "identity": 0.5071344375610352,
      "voiceclap": 0.04259336739778519,
      "fitness": 0.6479463934281552,
      "legacy_fitness": 0.6479463934281552,
      "target_cdf": 0.7466748362351533,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.828540802001953,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7691833218552662,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.828540802001953,
          "normalized": 0.7466748362351533,
          "contribution": 0.5226723853646073
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8837981224060059,
          "normalized": 0.8217031216355299,
          "contribution": 0.24651093649065897
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/46/46769a51d397a5a603d2a54d66da.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "40f14c623a531a73da655ba9a2c35a513e42516bdf212ef73ca647e188d22aaa",
      "id": "f1840e326aa39898136dcdb0a094",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 2,
      "generation": null,
      "ce": 5.796247482299805,
      "genuineness": 4.6203155517578125,
      "wer": 0.0,
      "identity": 0.6067171692848206,
      "voiceclap": 0.042130596935749054,
      "fitness": 0.688607946653267,
      "legacy_fitness": 0.688607946653267,
      "target_cdf": 0.702377223211726,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.6162185668945312,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7446323769104093,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.6162185668945312,
          "normalized": 0.702377223211726,
          "contribution": 0.4916640562482082
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8091652393341064,
          "normalized": 0.8432277355406702,
          "contribution": 0.25296832066220104
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/f1/f1840e326aa39898136dcdb0a094.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "f21e9a0e4902ddaf5b9351aaa8c965a3f36ddbbd8d3e75c41438893b5e89152c",
      "id": "63deaef30a20849e3a27e8f8ac40",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 2,
      "generation": null,
      "ce": 5.589737892150879,
      "genuineness": 4.337893962860107,
      "wer": 0.03225806451612903,
      "identity": 0.5146161317825317,
      "voiceclap": 0.02124226838350296,
      "fitness": 0.6677878615842384,
      "legacy_fitness": 0.6677878615842384,
      "target_cdf": 0.73911685576154,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.791072368621826,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7725654541234187,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.791072368621826,
          "normalized": 0.73911685576154,
          "contribution": 0.5173817990330779
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7820258140563965,
          "normalized": 0.8506121836344694,
          "contribution": 0.2551836550903408
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/63/63deaef30a20849e3a27e8f8ac40.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "b6307c8bcdf92bdb5f40cd1d2f7889550a797aa087a9b97f551a1d3800895ba0",
      "id": "030fc249f3a88697b67af2717c41",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 2,
      "generation": null,
      "ce": 5.685050010681152,
      "genuineness": 4.425758361816406,
      "wer": 0.03225806451612903,
      "identity": 0.542634129524231,
      "voiceclap": 0.056930214166641235,
      "fitness": 0.6169321571151469,
      "legacy_fitness": 0.6169321571151469,
      "target_cdf": 0.6219274273525455,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.2633888721466064,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6937231977227512,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.2633888721466064,
          "normalized": 0.6219274273525455,
          "contribution": 0.43534919914678183
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7413170337677002,
          "normalized": 0.8612466619198977,
          "contribution": 0.2583739985759693
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/03/030fc249f3a88697b67af2717c41.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "eb905e3d9cf920c3d5a09b1a3431294b85f968e0390a3543e57bf2959ff9c655",
      "id": "58411975a80ad230c8311ab21f03",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 2,
      "generation": null,
      "ce": 5.657250881195068,
      "genuineness": 4.344170093536377,
      "wer": 0.0967741935483871,
      "identity": 0.586076557636261,
      "voiceclap": 0.07457491010427475,
      "fitness": 0.5956520668157719,
      "legacy_fitness": 0.5956520668157719,
      "target_cdf": 0.6740671437138053,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.4882006645202637,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7264834585022412,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.4882006645202637,
          "normalized": 0.6740671437138053,
          "contribution": 0.47184700059966367
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7888107299804688,
          "normalized": 0.8487881930085917,
          "contribution": 0.2546364579025775
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/58/58411975a80ad230c8311ab21f03.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "7805bda791a7376fdda95918fabf295bbc7547ac67f217efed8310b35c84be49",
      "id": "2c16d9e38b3ba8949aac98b5ef29",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.71682071685791,
      "genuineness": 4.375550270080566,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.04343748837709427,
      "fitness": 0.5643319738050401,
      "legacy_fitness": 0.5643319738050401,
      "target_cdf": 0.4781937225938042,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.6795027256011963,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5642519346261536,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.6795027256011963,
          "normalized": 0.4781937225938042,
          "contribution": 0.3347356058156629
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0568110942840576,
          "normalized": 0.7650544293683025,
          "contribution": 0.22951632881049072
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/2c/2c16d9e38b3ba8949aac98b5ef29.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "13ecec8bb8ab44ece91a0446dcc8a94b62cfa03d0ad8b5857cd5d2863b24883d",
      "id": "d48061dcbec94e96b0ae2573eb93",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 2,
      "generation": null,
      "ce": 6.300607204437256,
      "genuineness": 4.632867813110352,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.002691030502319336,
      "fitness": 0.5576429962239372,
      "legacy_fitness": 0.5576429962239372,
      "target_cdf": 0.4141833144668156,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.420344591140747,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5683234291187659,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.420344591140747,
          "normalized": 0.4141833144668156,
          "contribution": 0.2899283201267709
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.415645718574524,
          "normalized": 0.9279836966399834,
          "contribution": 0.278395108991995
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d4/d48061dcbec94e96b0ae2573eb93.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "ec60fc459c8c7c9979999ddf20cf893458f2181080ae76e8bb672a7ff226a5f2",
      "id": "fbfcb0b9517abbb314ee9ec18602",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.387199878692627,
      "genuineness": 4.319066047668457,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.06002850458025932,
      "fitness": 0.6311475384259081,
      "legacy_fitness": 0.6311475384259081,
      "target_cdf": 0.6330129122781467,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.310224771499634,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7130931574607198,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.310224771499634,
          "normalized": 0.6330129122781467,
          "contribution": 0.4431090385947027
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.5716965198516846,
          "normalized": 0.8999470628867237,
          "contribution": 0.2699841188660171
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/fb/fbfcb0b9517abbb314ee9ec18602.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "1ebfd938f2b65f95c847a02b9ba587617db7e7ada4289fbc653212c5f70a3bee",
      "id": "4893e6ba14e167576ac6264acd46",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.68,
      "cfg": 2,
      "generation": null,
      "ce": 5.454712867736816,
      "genuineness": 4.369274139404297,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.03105328232049942,
      "fitness": 0.6044547321850873,
      "legacy_fitness": 0.6044547321850873,
      "target_cdf": 0.5720676426996423,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.057311534881592,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6661081548595614,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.057311534881592,
          "normalized": 0.5720676426996423,
          "contribution": 0.40044734988974956
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6395447254180908,
          "normalized": 0.8855360165660395,
          "contribution": 0.26566080496981187
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/48/4893e6ba14e167576ac6264acd46.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "92ef48874460ad19c0858865517a646bcbed69e6d31bcd71d181f4ad2178ef29",
      "id": "85ef630c025afd05e1bdbb9fb491",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.7247633934021,
      "genuineness": 4.65797233581543,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.022768981754779816,
      "fitness": 0.6120703723746677,
      "legacy_fitness": 0.6120703723746677,
      "target_cdf": 0.5497382137612778,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.9667621850967407,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6354988813750292,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.9667621850967407,
          "normalized": 0.5497382137612778,
          "contribution": 0.3848167496328945
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8363044261932373,
          "normalized": 0.8356071058071158,
          "contribution": 0.2506821317421347
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/85/85ef630c025afd05e1bdbb9fb491.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "252ae27d7868ee764c453e66bce2ba77ac83e0b3455defdc01267618707c77a6",
      "id": "cd15ba6ef3ee238af688a34058a6",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.3792572021484375,
      "genuineness": 4.218649387359619,
      "wer": 0.07407407407407407,
      "identity": 0.38099032640457153,
      "voiceclap": 0.04520454257726669,
      "fitness": 0.5421426993575447,
      "legacy_fitness": 0.5421426993575447,
      "target_cdf": 0.5505112548915733,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.9698846340179443,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.649170795403836,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.9698846340179443,
          "normalized": 0.5505112548915733,
          "contribution": 0.38535787842410124
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6666839122772217,
          "normalized": 0.8793763899324494,
          "contribution": 0.2638129169797348
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/cd/cd15ba6ef3ee238af688a34058a6.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "b5704e59f967905b3c334f44ff48863206c9c7a8bf8600bc9077e620a73fb86d",
      "id": "29319412d0cc2f6cffe6fd02f7df",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.7128496170043945,
      "genuineness": 4.362998008728027,
      "wer": 0.0,
      "identity": 0.44920217990875244,
      "voiceclap": 0.04004032909870148,
      "fitness": 0.644784175952864,
      "legacy_fitness": 0.644784175952864,
      "target_cdf": 0.6403427036615059,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.3414485454559326,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7120528095427889,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.3414485454559326,
          "normalized": 0.6403427036615059,
          "contribution": 0.44823989256305413
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6666839122772217,
          "normalized": 0.8793763899324494,
          "contribution": 0.2638129169797348
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/29/29319412d0cc2f6cffe6fd02f7df.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "ac1f999627b390b0a8e16e51eb7bae5c828c3a6e4f9e6ac8b0c1f31aa50e7f38",
      "id": "9a30e3538763f4e1e90ed4a16c3d",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.434855937957764,
      "genuineness": 4.413206577301025,
      "wer": 0.0,
      "identity": 0.28905054926872253,
      "voiceclap": 0.02542996034026146,
      "fitness": 0.5755980492594197,
      "legacy_fitness": 0.5755980492594197,
      "target_cdf": 0.5116860868458657,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.813765287399292,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6155084589153217,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.813765287399292,
          "normalized": 0.5116860868458657,
          "contribution": 0.35818026079210596
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7548866271972656,
          "normalized": 0.8577606604107191,
          "contribution": 0.2573281981232157
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/9a/9a30e3538763f4e1e90ed4a16c3d.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "4517960e7138ebd39cb7778786d9e0cc1e14df6f96a420682a96b370ca713c69",
      "id": "e4b42777a766fdd6a05afdea8ebc",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.863759994506836,
      "genuineness": 4.319066047668457,
      "wer": 0.07407407407407407,
      "identity": 0.49839407205581665,
      "voiceclap": 0.06145971268415451,
      "fitness": 0.5468230097355374,
      "legacy_fitness": 0.5468230097355374,
      "target_cdf": 0.5280275306643809,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.8793354034423828,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6352800764348785,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.8793354034423828,
          "normalized": 0.5280275306643809,
          "contribution": 0.3696192714650666
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6395447254180908,
          "normalized": 0.8855360165660395,
          "contribution": 0.26566080496981187
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/e4/e4b42777a766fdd6a05afdea8ebc.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "70fe6429c01e4376ae232183f6236a0b00abef3e3c2fc0d98e2a07bde88a3531",
      "id": "b3e5b33fedaa2f25f3356c5ade2f",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.6254801750183105,
      "genuineness": 4.469690799713135,
      "wer": 0.1111111111111111,
      "identity": 0.5664385557174683,
      "voiceclap": 0.0486256442964077,
      "fitness": 0.552269630543278,
      "legacy_fitness": 0.552269630543278,
      "target_cdf": 0.5888584266620323,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.1260039806365967,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6927587703789426,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.1260039806365967,
          "normalized": 0.5888584266620323,
          "contribution": 0.4122008986634226
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.3681519031524658,
          "normalized": 0.9351929057184,
          "contribution": 0.28055787171552
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/b3/b3e5b33fedaa2f25f3356c5ade2f.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "69b7f81d0c7f514d97bfc0606172c7372b78af89c8c596d28d59be76f2c6900c",
      "id": "04b6c6a54d3c6c0db10d64364163",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.523001670837402,
      "genuineness": 4.545003414154053,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.029724273830652237,
      "fitness": 0.5080968854267619,
      "legacy_fitness": 0.5080968854267619,
      "target_cdf": 0.31129340279881595,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.9800881743431091,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5001798464425129,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.9800881743431091,
          "normalized": 0.31129340279881595,
          "contribution": 0.21790538195917117
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.32744300365448,
          "normalized": 0.9409148816111391,
          "contribution": 0.2822744644833417
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/04/04b6c6a54d3c6c0db10d64364163.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "b4ffed5573c3f220024f38e987c5eb214ac3f803185f958ad1e9e9ecc028c094",
      "id": "eae208baa828e683a6ba8d41b48f",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.38003396987915,
      "genuineness": 4.394378662109375,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.012686258181929588,
      "fitness": 0.4630944219162638,
      "legacy_fitness": 0.4630944219162638,
      "target_cdf": 0.24098892349612216,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.6428704261779785,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4291051283999292,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.6428704261779785,
          "normalized": 0.24098892349612216,
          "contribution": 0.1686922464472855
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7141776084899902,
          "normalized": 0.8680429398421459,
          "contribution": 0.2604128819526437
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/ea/eae208baa828e683a6ba8d41b48f.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "93d80376fe96a84e10d1003f4c2cab244fd9f41af5e45c36d2a4051cd2982df3",
      "id": "f7e6edd2dd23f0476f71eadff6f6",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.3641486167907715,
      "genuineness": 4.469690799713135,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.11304350197315216,
      "fitness": 0.4545798283156744,
      "legacy_fitness": 0.4545798283156744,
      "target_cdf": 0.2184779924823823,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.5242197513580322,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.43289243302363634,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.5242197513580322,
          "normalized": 0.2184779924823823,
          "contribution": 0.1529345947376676
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.3817216157913208,
          "normalized": 0.9331927942865623,
          "contribution": 0.2799578382859687
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/f7/f7e6edd2dd23f0476f71eadff6f6.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "e5f8429eeb94ab15f93a027b6847be6db371b3d169059d757fafd6b5760a6f9c",
      "id": "22ba215ecd860e933f67635603d0",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.443574905395508,
      "genuineness": 4.519899368286133,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.029902763664722443,
      "fitness": 0.46097548461621796,
      "legacy_fitness": 0.46097548461621796,
      "target_cdf": 0.22311394327214934,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.5491988658905029,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4140330699515471,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.5491988658905029,
          "normalized": 0.22311394327214934,
          "contribution": 0.15617976029050454
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7481017112731934,
          "normalized": 0.8595110322034752,
          "contribution": 0.2578533096610426
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/22/22ba215ecd860e933f67635603d0.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "281fbd0c6378a8aadb011c0a8b32c759d49790e8a741a1493fd54c85ef2923d2",
      "id": "6bfb08eda616f8b2507bb6382798",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.475345611572266,
      "genuineness": 4.21237325668335,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.05383335426449776,
      "fitness": 0.43655655195598386,
      "legacy_fitness": 0.43655655195598386,
      "target_cdf": 0.1983147186097421,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.4118138551712036,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4002263561500601,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.4118138551712036,
          "normalized": 0.1983147186097421,
          "contribution": 0.13882030302681944
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7006080150604248,
          "normalized": 0.8713535104108022,
          "contribution": 0.26140605312324067
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/6b/6bfb08eda616f8b2507bb6382798.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "4464ed8b1a72628d1074de87836668820534281ec4393e4dee23e64e3ac37fd2",
      "id": "6cf0d4234bae1830f3c2a23867de",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.372091293334961,
      "genuineness": 4.2249250411987305,
      "wer": 0.0,
      "identity": 0.7451968789100647,
      "voiceclap": 0.14985689520835876,
      "fitness": 0.4339550651006199,
      "legacy_fitness": 0.4339550651006199,
      "target_cdf": 0.19722847876793087,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.40556907653808594,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.41740351628375516,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.40556907653808594,
          "normalized": 0.19722847876793087,
          "contribution": 0.1380599351375516
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.3952912092208862,
          "normalized": 0.9311452704873452,
          "contribution": 0.27934358114620356
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/6c/6cf0d4234bae1830f3c2a23867de.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "420887269ec933a9595249f259e0c5dc11b41a307a7fe41c5fdf129e40b4ade0",
      "id": "960e6efe8d17eaa24b2989f5d9fb",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.427689552307129,
      "genuineness": 4.344170093536377,
      "wer": 0.0,
      "identity": 0.6330074071884155,
      "voiceclap": 0.052913542836904526,
      "fitness": 0.46219378778642173,
      "legacy_fitness": 0.46219378778642173,
      "target_cdf": 0.24098892349612216,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.6428704261779785,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4320324408408587,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.6428704261779785,
          "normalized": 0.24098892349612216,
          "contribution": 0.1686922464472855
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.673468828201294,
          "normalized": 0.8778006479785774,
          "contribution": 0.2633401943935732
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/96/960e6efe8d17eaa24b2989f5d9fb.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "234c872cd1d39f775432d2e4720bddde5603e1e66e6857ca7d080e8cda3a15ba",
      "id": "d72b19b78cc25f1903c55725625d",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.403861999511719,
      "genuineness": 4.419482707977295,
      "wer": 0.0,
      "identity": 0.6991642713546753,
      "voiceclap": 0.10271569341421127,
      "fitness": 0.5270002466137822,
      "legacy_fitness": 0.5270002466137822,
      "target_cdf": 0.365517167587204,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.2173895835876465,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5219741098374268,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.2173895835876465,
          "normalized": 0.365517167587204,
          "contribution": 0.2558620173110428
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.632759928703308,
          "normalized": 0.8870403084212799,
          "contribution": 0.26611209252638396
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d7/d72b19b78cc25f1903c55725625d.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "fdfbf3c5a71205f7ac00218d09ce52f395066b25d18b633c9912f6cf7cff1689",
      "id": "58474e133d2f60a313c76fb4c115",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.340320587158203,
      "genuineness": 4.432034492492676,
      "wer": 0.0,
      "identity": 0.7737911939620972,
      "voiceclap": 0.10413192212581635,
      "fitness": 0.49744488789581004,
      "legacy_fitness": 0.49744488789581004,
      "target_cdf": 0.3085375387259869,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.9675986170768738,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.48835575330391545,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.9675986170768738,
          "normalized": 0.3085375387259869,
          "contribution": 0.2159762771081908
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.5309876203536987,
          "normalized": 0.9079315873190821,
          "contribution": 0.2723794761957246
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/58/58474e133d2f60a313c76fb4c115.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "b4c359ffb543f2109c2122e6859d62f9b5713f80a73351b52b06eeffbbc52f0c",
      "id": "d5c2a3018b9edb6d8cc6e7a591e8",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "The last light in the building went out a minute ago. I was still here... trying to remember the question you asked. Let me answer it before we leave this place.",
      "sentences": [
        "The last light in the building went out a minute ago.",
        "I was still here... trying to remember the question you asked.",
        "Let me answer it before we leave this place."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] The last light in the building went out a minute ago.\n(Natural clear speech.) [4.2 seconds duration] I was still here... trying to remember the question you asked.\n(Natural clear speech.) [3.5 seconds duration] Let me answer it before we leave this place.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.340320587158203,
      "genuineness": 4.532451152801514,
      "wer": 0.0,
      "identity": 0.5741070508956909,
      "voiceclap": 0.089301198720932,
      "fitness": 0.5884025399763589,
      "legacy_fitness": 0.5884025399763589,
      "target_cdf": 0.48208478786134823,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.6951146125793457,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6074434703689608,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.6951146125793457,
          "normalized": 0.48208478786134823,
          "contribution": 0.33745935150294376
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.5716965198516846,
          "normalized": 0.8999470628867237,
          "contribution": 0.2699841188660171
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d5/d5c2a3018b9edb6d8cc6e7a591e8.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "bc4b46fa4280154184cfc221faeab435629f4669a219cf4020fe0ae6654cf898",
      "id": "74109b4f05d559c9798253207085",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.546830177307129,
      "genuineness": 4.557555198669434,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.036387793719768524,
      "fitness": 0.4538178884958342,
      "legacy_fitness": 0.4538178884958342,
      "target_cdf": 0.20049800157052589,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.42430341243743896,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.3874686664723974,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.42430341243743896,
          "normalized": 0.20049800157052589,
          "contribution": 0.14034860109936811
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8770133256912231,
          "normalized": 0.8237335512434312,
          "contribution": 0.24712006537302933
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/74/74109b4f05d559c9798253207085.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "8728e2f4a65fd7db987e0b2d645697ba9d8114c9cb8c123ba694073abb96597f",
      "id": "5eb74bc225f36c6f688369f1d966",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.4515180587768555,
      "genuineness": 4.331617832183838,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.12056265026330948,
      "fitness": 0.5282151317014051,
      "legacy_fitness": 0.5282151317014051,
      "target_cdf": 0.37288620778198095,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.2486133575439453,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5114117153847071,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.2486133575439453,
          "normalized": 0.37288620778198095,
          "contribution": 0.26102034544738667
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8396968841552734,
          "normalized": 0.8346378997910682,
          "contribution": 0.2503913699373204
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/5e/5eb74bc225f36c6f688369f1d966.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "18f8e4d9af42c4134b031daa214ce7ae4c162b88b7ea48312de0019e2660ee77",
      "id": "adb9f88e1cd6f62826dba70bcbff",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.507116317749023,
      "genuineness": 4.369274139404297,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.0228513702750206,
      "fitness": 0.5958997603059326,
      "legacy_fitness": 0.5958997603059326,
      "target_cdf": 0.5023375264407227,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.776296615600586,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6020276384458263,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.776296615600586,
          "normalized": 0.5023375264407227,
          "contribution": 0.3516362685085059
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8396968841552734,
          "normalized": 0.8346378997910682,
          "contribution": 0.2503913699373204
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/ad/adb9f88e1cd6f62826dba70bcbff.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "389de22586b8e77ec7f3039271186deafcd7224a4e4c61f029eec57d2ee4751f",
      "id": "8302e6a95828bb7788ab214c252e",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.372091293334961,
      "genuineness": 4.406930446624756,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.051769234240055084,
      "fitness": 0.5398046158061491,
      "legacy_fitness": 0.5398046158061491,
      "target_cdf": 0.39376046306015383,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.3360402584075928,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5018671670013862,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.3360402584075928,
          "normalized": 0.39376046306015383,
          "contribution": 0.27563232414210764
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0873427391052246,
          "normalized": 0.7541161428642621,
          "contribution": 0.2262348428592786
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/83/8302e6a95828bb7788ab214c252e.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "b158d0f7781273bd954e88456c8ea51f1c082447a9205ade44cca39f9c30cfa4",
      "id": "d53dca1ea3ac99b5fa02c8cbe50a",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.610371112823486,
      "genuineness": 4.406930446624756,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.028235414996743202,
      "fitness": 0.4355055554728479,
      "legacy_fitness": 0.4355055554728479,
      "target_cdf": 0.1732483514191252,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.26193928718566895,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.373107932465391,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.26193928718566895,
          "normalized": 0.1732483514191252,
          "contribution": 0.12127384599338763
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8227348327636719,
          "normalized": 0.8394469549066778,
          "contribution": 0.25183408647200334
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d5/d53dca1ea3ac99b5fa02c8cbe50a.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "fe25f2f4ca6f9a4cbb5a8511756acfe41fd838a94e8a8ffc247ea8241c8e730d",
      "id": "fbdef858edd42a823a1fa2fb792a",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.395919322967529,
      "genuineness": 4.143336772918701,
      "wer": 0.0,
      "identity": 0.5662573575973511,
      "voiceclap": 0.08934707194566727,
      "fitness": 0.4333345328251379,
      "legacy_fitness": 0.4333345328251379,
      "target_cdf": 0.2015950350920075,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.43054819107055664,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4094215896438108,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.43054819107055664,
          "normalized": 0.2015950350920075,
          "contribution": 0.14111652456440524
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.598835825920105,
          "normalized": 0.894350216931352,
          "contribution": 0.2683050650794056
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/fb/fbdef858edd42a823a1fa2fb792a.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "5eb3d4a8bafd2dd28ccab36a662b3c9ba1028100c068115eace5854873d1f245",
      "id": "0ededad076b492ba47c32e10fbea",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.324435234069824,
      "genuineness": 4.406930446624756,
      "wer": 0.0,
      "identity": 0.6902761459350586,
      "voiceclap": 0.08442476391792297,
      "fitness": 0.5442701153976737,
      "legacy_fitness": 0.5442701153976737,
      "target_cdf": 0.40507426520645995,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.382875919342041,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.530974957399507,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.382875919342041,
          "normalized": 0.40507426520645995,
          "contribution": 0.28355198564452194
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8736209869384766,
          "normalized": 0.8247432391832834,
          "contribution": 0.247422971754985
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/0ededad076b492ba47c32e10fbea.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "85370fc0ab2da878b47c62c9124f6a18b4928ffcb04b115e29a9ee4ce98aecec",
      "id": "73f0280302595440100991a1df59",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.340320587158203,
      "genuineness": 4.331617832183838,
      "wer": 0.0,
      "identity": 0.5562503933906555,
      "voiceclap": 0.09027722477912903,
      "fitness": 0.5514265142007719,
      "legacy_fitness": 0.5514265142007719,
      "target_cdf": 0.4248688463616471,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.4640580415725708,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5673923113191701,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.4640580415725708,
          "normalized": 0.4248688463616471,
          "contribution": 0.297408192453153
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.5716965198516846,
          "normalized": 0.8999470628867237,
          "contribution": 0.2699841188660171
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/73/73f0280302595440100991a1df59.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "c437d0feb841f51dee3135bbb6c0fb570b93f2dfd1b4d637027f3f212de799a0",
      "id": "e9c0015a69ec688b39585402e6ee",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.499174118041992,
      "genuineness": 4.3127899169921875,
      "wer": 0.0,
      "identity": 0.5627975463867188,
      "voiceclap": 0.12512671947479248,
      "fitness": 0.4849583463161095,
      "legacy_fitness": 0.4849583463161095,
      "target_cdf": 0.2855588269807704,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.8614374995231628,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.47892223803506373,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.8614374995231628,
          "normalized": 0.2855588269807704,
          "contribution": 0.19989117888653926
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.402076005935669,
          "normalized": 0.9301035304950818,
          "contribution": 0.2790310591485245
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/e9/e9c0015a69ec688b39585402e6ee.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "559474aaa118d59308ae06895b47984c4f133780f9c82c38a8d996f3c12f16f8",
      "id": "83d128d4708075913b502c32a331",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Someone left your umbrella beside the narrow garden gate. I brought it inside... and then I noticed what was underneath it. You should look at this before I try to explain.",
      "sentences": [
        "Someone left your umbrella beside the narrow garden gate.",
        "I brought it inside... and then I noticed what was underneath it.",
        "You should look at this before I try to explain."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] Someone left your umbrella beside the narrow garden gate.\n(Natural clear speech.) [4.6 seconds duration] I brought it inside... and then I noticed what was underneath it.\n(Natural clear speech.) [3.8 seconds duration] You should look at this before I try to explain.",
      "requested_seconds": 11.92,
      "actual_seconds": 11.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.3641486167907715,
      "genuineness": 4.607763767242432,
      "wer": 0.0,
      "identity": 0.4251393973827362,
      "voiceclap": 0.05796879529953003,
      "fitness": 0.6081050921595493,
      "legacy_fitness": 0.6081050921595493,
      "target_cdf": 0.5140224395426908,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.8231325149536133,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6075404937849013,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.8231325149536133,
          "normalized": 0.5140224395426908,
          "contribution": 0.3598157076798835
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8702285289764404,
          "normalized": 0.8257492870167258,
          "contribution": 0.24772478610501772
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/83/83d128d4708075913b502c32a331.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "2699603f0ea53b941070b85058db09c95aeab1dca3e127c5ce865cc7b770be2f",
      "id": "f0c79ed0c387952fc1edb7e9fec6",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.483288764953613,
      "genuineness": 4.325342178344727,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.10448876023292542,
      "fitness": 0.5339452977103872,
      "legacy_fitness": 0.5339452977103872,
      "target_cdf": 0.38328097564436653,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.292326807975769,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5406761591467812,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.292326807975769,
          "normalized": 0.38328097564436653,
          "contribution": 0.2682966829510566
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.5309876203536987,
          "normalized": 0.9079315873190821,
          "contribution": 0.2723794761957246
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/f0/f0c79ed0c387952fc1edb7e9fec6.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "c007ea69b94ce04eccea5e484277635795ed21705ae3011325d188ca5b4d70a0",
      "id": "3f750438738f478f61038204c6a8",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.475345611572266,
      "genuineness": 4.444586753845215,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.04446412995457649,
      "fitness": 0.5364823988048718,
      "legacy_fitness": 0.5364823988048718,
      "target_cdf": 0.37881528754402916,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.273592472076416,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.49718250021804034,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.273592472076416,
          "normalized": 0.37881528754402916,
          "contribution": 0.2651707012808204
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.033064126968384,
          "normalized": 0.7733726631240666,
          "contribution": 0.23201179893721996
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/3f/3f750438738f478f61038204c6a8.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "8b53306678880b9665582bcdd09903816c54428b7f602d97a0cba0fe1cf11ca5",
      "id": "1ab7647a0a24a0f5c12492e2abfc",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.507116317749023,
      "genuineness": 4.557555198669434,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.06486229598522186,
      "fitness": 0.5442165290438019,
      "legacy_fitness": 0.5442165290438019,
      "target_cdf": 0.38328097564436653,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.292326807975769,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5010121872642019,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.292326807975769,
          "normalized": 0.38328097564436653,
          "contribution": 0.2682966829510566
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0262794494628906,
          "normalized": 0.7757183477104845,
          "contribution": 0.23271550431314536
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/1a/1ab7647a0a24a0f5c12492e2abfc.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "5adee4e37c3ebfeae2c222d2ac24eec0cc6a4e9ab3e5a415e2a2ea2c4d9b9e45",
      "id": "ad55f397ae614368da380b0e5b36",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.515059471130371,
      "genuineness": 4.582659721374512,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.05332805961370468,
      "fitness": 0.4792364642587297,
      "legacy_fitness": 0.4792364642587297,
      "target_cdf": 0.25083164484639814,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.6928286552429199,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.42209308788313765,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.6928286552429199,
          "normalized": 0.25083164484639814,
          "contribution": 0.17558215139247868
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8837981224060059,
          "normalized": 0.8217031216355299,
          "contribution": 0.24651093649065897
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/ad/ad55f397ae614368da380b0e5b36.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "7d1bbf571099308ca299a32066fea14c3759a61c66f6500f25da2f772b139958",
      "id": "2c6d24d17b5e10ec3cd3cc7352ee",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.507116317749023,
      "genuineness": 4.381826400756836,
      "wer": 0.030303030303030304,
      "identity": null,
      "voiceclap": 0.09478403627872467,
      "fitness": 0.5389351831522249,
      "legacy_fitness": 0.5389351831522249,
      "target_cdf": 0.42104579930094294,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.4484460353851318,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5504584687637379,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.4484460353851318,
          "normalized": 0.42104579930094294,
          "contribution": 0.29473205951066006
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7752411365509033,
          "normalized": 0.8524213641769265,
          "contribution": 0.25572640925307794
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/2c/2c6d24d17b5e10ec3cd3cc7352ee.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "8777f5d671722f362dd7396077400c04e4177798b6aca891a6ae1a3e462fed70",
      "id": "3d75793338e5086a8fe8002b12a3",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.435632705688477,
      "genuineness": 4.595211505889893,
      "wer": 0.0,
      "identity": 0.6547054648399353,
      "voiceclap": 0.07150068879127502,
      "fitness": 0.5678581023210995,
      "legacy_fitness": 0.5678581023210995,
      "target_cdf": 0.43100027720028417,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.4890371561050415,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5308559372475342,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.4890371561050415,
          "normalized": 0.43100027720028417,
          "contribution": 0.3017001940401989
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0602035522460938,
          "normalized": 0.7638524773577846,
          "contribution": 0.22915574320733537
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/3d/3d75793338e5086a8fe8002b12a3.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "e4f2293ba1184ee5552d4e2b8c7b6c2611381b93a9fe0f77f9c2883b454bfb21",
      "id": "bf0d3ddea265362e0597bf53a27c",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.499174118041992,
      "genuineness": 4.394378662109375,
      "wer": 0.0,
      "identity": 0.5760014653205872,
      "voiceclap": 0.1306716352701187,
      "fitness": 0.5718694496116545,
      "legacy_fitness": 0.5718694496116545,
      "target_cdf": 0.4525819714787615,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.5764639377593994,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5815528123081133,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.5764639377593994,
          "normalized": 0.4525819714787615,
          "contribution": 0.31680738003513303
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6531143188476562,
          "normalized": 0.8824847742432673,
          "contribution": 0.2647454322729802
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/bf/bf0d3ddea265362e0597bf53a27c.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "6aea26b5d7adb1dd58946240eb7fa67eac1bfc01613512e66271f35f113cfc8d",
      "id": "1f0c0b7eb9acdb44b4c88f0effae",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.340320587158203,
      "genuineness": 4.111956596374512,
      "wer": 0.0,
      "identity": 0.38171350955963135,
      "voiceclap": 0.1632189154624939,
      "fitness": 0.4779749941299464,
      "legacy_fitness": 0.4779749941299464,
      "target_cdf": 0.2962709092041066,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.9113956689834595,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.48344215090995785,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.9113956689834595,
          "normalized": 0.2962709092041066,
          "contribution": 0.20738963644287461
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.4631394147872925,
          "normalized": 0.9201750482236108,
          "contribution": 0.27605251446708323
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/1f/1f0c0b7eb9acdb44b4c88f0effae.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "adf2e98796ae131427f6bc19282504b2039bac00f4a3ce16e4030188ac2a89aa",
      "id": "b9956c9cf3efc46d3a052605ac14",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.475345611572266,
      "genuineness": 4.168440818786621,
      "wer": 0.0,
      "identity": 0.6227669715881348,
      "voiceclap": 0.15931418538093567,
      "fitness": 0.4737385055688669,
      "legacy_fitness": 0.4737385055688669,
      "target_cdf": 0.2763396623269021,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.8177240490913391,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4705167565886423,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.8177240490913391,
          "normalized": 0.2763396623269021,
          "contribution": 0.19343776362883147
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.4427849054336548,
          "normalized": 0.9235966431993694,
          "contribution": 0.2770789929598108
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/b9/b9956c9cf3efc46d3a052605ac14.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "cab5d5bede3113da7398fe0ca8082d031ede9befbd208eeb547a9ee4f4b3e7a1",
      "id": "d428132f4b298b2c3c86d9018d32",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "I have kept this small box since the day we met. There is nothing expensive inside... only something I could not throw away. Open it slowly and then tell me what you think.",
      "sentences": [
        "I have kept this small box since the day we met.",
        "There is nothing expensive inside... only something I could not throw away.",
        "Open it slowly and then tell me what you think."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "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 have kept this small box since the day we met.\n(Natural clear speech.) [4.6 seconds duration] There is nothing expensive inside... only something I could not throw away.\n(Natural clear speech.) [3.8 seconds duration] Open it slowly and then tell me what you think.",
      "requested_seconds": 12.64,
      "actual_seconds": 12.64,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.38797664642334,
      "genuineness": 4.2249250411987305,
      "wer": 0.0,
      "identity": 0.6431508660316467,
      "voiceclap": 0.09962242841720581,
      "fitness": 0.49000672890685737,
      "legacy_fitness": 0.49000672890685737,
      "target_cdf": 0.3085375387259869,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.9675986170768738,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4923747428521488,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.9675986170768738,
          "normalized": 0.3085375387259869,
          "contribution": 0.2159762771081908
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.4563546180725098,
          "normalized": 0.9213282191465266,
          "contribution": 0.276398465743958
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d4/d428132f4b298b2c3c86d9018d32.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "1d4584033d71c5497238c380e3debf8516009379e54cd84e5781138dda9f28c1",
      "id": "001cebd0a5f9054401af46e22945",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.181467056274414,
      "genuineness": 4.632867813110352,
      "wer": 0.0967741935483871,
      "identity": null,
      "voiceclap": 0.01987704262137413,
      "fitness": 0.4931228263073205,
      "legacy_fitness": 0.4931228263073205,
      "target_cdf": 0.39676915910757893,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.3485298156738281,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5230178874847502,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.3485298156738281,
          "normalized": 0.39676915910757893,
          "contribution": 0.27773841137530525
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8973677158355713,
          "normalized": 0.8175982536981496,
          "contribution": 0.2452794761094449
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/00/001cebd0a5f9054401af46e22945.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "e0c7cd33462722375258913c7931628822c66eb9fd74c2720cff2749f96ef140",
      "id": "fae24e9dd6a9bd2cc858e3dafb15",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.062326908111572,
      "genuineness": 4.65797233581543,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.029826033860445023,
      "fitness": 0.6356901811377094,
      "legacy_fitness": 0.6356901811377094,
      "target_cdf": 0.5800996555518876,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.0900964736938477,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5721501661898465,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.0900964736938477,
          "normalized": 0.5800996555518876,
          "contribution": 0.4060697588863213
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.567368745803833,
          "normalized": 0.5536013576784171,
          "contribution": 0.16608040730352513
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/fa/fae24e9dd6a9bd2cc858e3dafb15.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "0186610fc664ec4bffd02b8bba20e086d853fbff6794f9e665aef540872e2e2f",
      "id": "299a79b185d3030627bc98137a1a",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.1894097328186035,
      "genuineness": 4.770940780639648,
      "wer": 0.0967741935483871,
      "identity": null,
      "voiceclap": 0.03490031138062477,
      "fitness": 0.5475819975237032,
      "legacy_fitness": 0.5475819975237032,
      "target_cdf": 0.5054541094892037,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.7887861728668213,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.600022591883787,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.7887861728668213,
          "normalized": 0.5054541094892037,
          "contribution": 0.3538178766424425
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.887190580368042,
          "normalized": 0.820682384137815,
          "contribution": 0.2462047152413445
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/29/299a79b185d3030627bc98137a1a.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "635ca7a6b89ba20dfba195827c7775994d3e84a6ac3c196237904d5cde9b6e34",
      "id": "e415fc9bb96eca5573b78d131e95",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.197352409362793,
      "genuineness": 4.582659721374512,
      "wer": 0.06451612903225806,
      "identity": null,
      "voiceclap": 0.013948324136435986,
      "fitness": 0.5298368693618422,
      "legacy_fitness": 0.5298368693618422,
      "target_cdf": 0.44099875563918695,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.5296281576156616,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5947464070867576,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.5296281576156616,
          "normalized": 0.44099875563918695,
          "contribution": 0.30869912894743085
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.2256708145141602,
          "normalized": 0.9534909271310894,
          "contribution": 0.2860472781393268
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/e4/e415fc9bb96eca5573b78d131e95.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "88638ebfe55cdbf97e81decda962b3ab4144180b27532ad08f2532bd7cdb2d30",
      "id": "dab1000556c0d7f7d44e83f80a8f",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.221180438995361,
      "genuineness": 4.720732688903809,
      "wer": 0.0967741935483871,
      "identity": null,
      "voiceclap": 0.0166904479265213,
      "fitness": 0.5688510256694451,
      "legacy_fitness": 0.5688510256694451,
      "target_cdf": 0.5551452870048766,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.9886189699172974,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5668522708441098,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.9886189699172974,
          "normalized": 0.5551452870048766,
          "contribution": 0.3886017009034136
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.4774696826934814,
          "normalized": 0.5941685664689873,
          "contribution": 0.17825056994069619
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/da/dab1000556c0d7f7d44e83f80a8f.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "090cb7487caa3c92b093e970b85ee403473f5be68a567be47d737da2abb25a53",
      "id": "1c15e8b1cf76f6c62b7e9abc75d1",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 1.0,
      "generation": null,
      "ce": 5.903473377227783,
      "genuineness": 4.645420074462891,
      "wer": 0.0967741935483871,
      "identity": 0.6250973343849182,
      "voiceclap": 0.030452875420451164,
      "fitness": 0.6197835613799263,
      "legacy_fitness": 0.6197835613799263,
      "target_cdf": 0.6900858775127308,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.5600156784057617,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7393248757027939,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.5600156784057617,
          "normalized": 0.6900858775127308,
          "contribution": 0.4830601142589115
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.768456220626831,
          "normalized": 0.8542158714796083,
          "contribution": 0.2562647614438825
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/1c/1c15e8b1cf76f6c62b7e9abc75d1.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "c2178526a69cff4bd0a617740a9e989e06f0891c6170759ee3f0ff7c2fa0df37",
      "id": "9fbe18801d6054355126b4672109",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.23706579208374,
      "genuineness": 4.821148872375488,
      "wer": 0.0967741935483871,
      "identity": 0.6451568603515625,
      "voiceclap": 0.013296498917043209,
      "fitness": 0.5603910638384256,
      "legacy_fitness": 0.5603910638384256,
      "target_cdf": 0.5272502314353218,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.8762130737304688,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6274491605806946,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.8762130737304688,
          "normalized": 0.5272502314353218,
          "contribution": 0.36907516200472523
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7413170337677002,
          "normalized": 0.8612466619198977,
          "contribution": 0.2583739985759693
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/9f/9fbe18801d6054355126b4672109.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "70eff83fe76a3aee8066907daec7c2afd3a5c1811c150e453aa6ef8d4349742f",
      "id": "b062e675e040d0ae737d3c80b3fa",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.24500846862793,
      "genuineness": 4.557555198669434,
      "wer": 0.06451612903225806,
      "identity": 0.324821412563324,
      "voiceclap": -0.054839059710502625,
      "fitness": 0.5444605745219557,
      "legacy_fitness": 0.5444605745219557,
      "target_cdf": 0.47197246933561904,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.6545236110687256,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4828349806852108,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.6545236110687256,
          "normalized": 0.47197246933561904,
          "contribution": 0.33038072853493333
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.6665966510772705,
          "normalized": 0.5081808405009249,
          "contribution": 0.15245425215027747
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/b0/b062e675e040d0ae737d3c80b3fa.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "20480893104d13f3384dfb7f5815a7cc8deff173024fdb6b89612b298c5f7ce0",
      "id": "00c99d30a0ac87371386f25bcac9",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.141753673553467,
      "genuineness": 4.7081804275512695,
      "wer": 0.06451612903225806,
      "identity": 0.18122598528862,
      "voiceclap": -0.023132598027586937,
      "fitness": 0.5479168893113073,
      "legacy_fitness": 0.5479168893113073,
      "target_cdf": 0.47197246933561904,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.6545236110687256,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5272660700639666,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.6545236110687256,
          "normalized": 0.47197246933561904,
          "contribution": 0.33038072853493333
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.334988594055176,
          "normalized": 0.6562844717634445,
          "contribution": 0.19688534152903334
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/00/00c99d30a0ac87371386f25bcac9.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "92f1a366159f0ade74f4d2ce2442a3b8c42f60e3175db172fcc8f2e3b3727b7a",
      "id": "81290fe0adba5fd5fee091ce31cb",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt. Aber da ist er schon... und ich habe dir noch etwas nicht gesagt. Bitte bleib bei mir, bis ich fertig bin.",
      "sentences": [
        "Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.",
        "Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.",
        "Bitte bleib bei mir, bis ich fertig bin."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "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 dachte, wir hätten mehr Zeit, bevor der Zug kommt.\n(Natural clear speech.) [5.5 seconds duration] Aber da ist er schon... und ich habe dir noch etwas nicht gesagt.\n(Natural clear speech.) [3.4 seconds duration] Bitte bleib bei mir, bis ich fertig bin.",
      "requested_seconds": 13.2,
      "actual_seconds": 13.2,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.165581703186035,
      "genuineness": 4.519899368286133,
      "wer": 0.06451612903225806,
      "identity": 0.26319706439971924,
      "voiceclap": -0.05207300931215286,
      "fitness": 0.5502361803097094,
      "legacy_fitness": 0.5502361803097094,
      "target_cdf": 0.49142964032956604,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.7325832843780518,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5841793820672363,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.7325832843780518,
          "normalized": 0.49142964032956604,
          "contribution": 0.3440007482306962
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.951646327972412,
          "normalized": 0.8005954461218004,
          "contribution": 0.24017863383654012
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/81/81290fe0adba5fd5fee091ce31cb.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "4bbe3f0ba945f92a2061c66a4cb9506119d682326db190bb39485c0398c884b2",
      "id": "c7df062cd918d6c822da61a44c2d",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.0305562019348145,
      "genuineness": 4.7332844734191895,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.007674764841794968,
      "fitness": 0.5777600732376948,
      "legacy_fitness": 0.5777600732376948,
      "target_cdf": 0.4595519635937164,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.6045653820037842,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5585363703854924,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.6045653820037842,
          "normalized": 0.4595519635937164,
          "contribution": 0.32168637451560145
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.9855704307556152,
          "normalized": 0.7894999862329698,
          "contribution": 0.23684999586989092
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/c7/c7df062cd918d6c822da61a44c2d.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "33046bced79425630a92e97c83b49e9e195951c5dfd56ddc36e0bc296fe031fa",
      "id": "a787d3986135e55fd775d49456ec",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.117925643920898,
      "genuineness": 4.720732688903809,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.0003241226077079773,
      "fitness": 0.5265604548656874,
      "legacy_fitness": 0.5265604548656874,
      "target_cdf": 0.35383023679334563,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.167431354522705,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4814444707715587,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.167431354522705,
          "normalized": 0.35383023679334563,
          "contribution": 0.24768116575534194
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0161020755767822,
          "normalized": 0.7792110167207227,
          "contribution": 0.23376330501621678
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/a7/a787d3986135e55fd775d49456ec.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "f28b629da4c069e799cf80b69b2c7e301db64970e06cd075fbad7da4c056ad08",
      "id": "d06abfd046972cc324e66b73d45b",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.38003396987915,
      "genuineness": 4.7458367347717285,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.005233835428953171,
      "fitness": 0.5466516903011487,
      "legacy_fitness": 0.5466516903011487,
      "target_cdf": 0.37881528754402916,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.273592472076416,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.526576754404061,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.273592472076416,
          "normalized": 0.37881528754402916,
          "contribution": 0.2651707012808204
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7006080150604248,
          "normalized": 0.8713535104108022,
          "contribution": 0.26140605312324067
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d0/d06abfd046972cc324e66b73d45b.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "5f04999d8b972e4d81c04c65dbdac045c0828b9f614f4d360682e8199e8cf01f",
      "id": "37c079824a371c698794ae15ce50",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.24500846862793,
      "genuineness": 4.959221839904785,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.014662478119134903,
      "fitness": 0.5732823546955358,
      "legacy_fitness": 0.5732823546955358,
      "target_cdf": 0.42104579930094294,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.4484460353851318,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5704348010438587,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.4484460353851318,
          "normalized": 0.42104579930094294,
          "contribution": 0.29473205951066006
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.4699242115020752,
          "normalized": 0.9190091384439952,
          "contribution": 0.27570274153319857
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/37/37c079824a371c698794ae15ce50.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "b9155aec4a61d625b9df515fd1682ad19ae61d979fb3263c4ed4855c6bbf8820",
      "id": "046d26a7e84f3078b7f02a9ac112",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.205295085906982,
      "genuineness": 4.595211505889893,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.014852209016680717,
      "fitness": 0.5533107249247924,
      "legacy_fitness": 0.5533107249247924,
      "target_cdf": 0.4134224033967445,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.417222261428833,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5593798012437382,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.417222261428833,
          "normalized": 0.4134224033967445,
          "contribution": 0.2893956823777211
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.5716965198516846,
          "normalized": 0.8999470628867237,
          "contribution": 0.2699841188660171
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/04/046d26a7e84f3078b7f02a9ac112.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "7fada64175cb38d0348eedea5d1ca363b67e7575c2039582df68feb0a5009fd9",
      "id": "966cdcc56d3930358c1c344220d7",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.316492557525635,
      "genuineness": 4.65797233581543,
      "wer": 0.0,
      "identity": 0.7160268425941467,
      "voiceclap": 0.0004948489367961884,
      "fitness": 0.5609908623897697,
      "legacy_fitness": 0.5609908623897697,
      "target_cdf": 0.4179927355853052,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.435956597328186,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5525046478934704,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.435956597328186,
          "normalized": 0.4179927355853052,
          "contribution": 0.2925949149097136
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7209625244140625,
          "normalized": 0.8663657766125228,
          "contribution": 0.2599097329837568
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/96/966cdcc56d3930358c1c344220d7.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "e1f8c40f62bc6674c204b287e9b2531bd524a4d4d041c6028f16b6f7566652b4",
      "id": "3bb22c9025660aa6dae32e778579",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.292664527893066,
      "genuineness": 4.68307638168335,
      "wer": 0.0,
      "identity": 0.516238808631897,
      "voiceclap": -0.013449788093566895,
      "fitness": 0.5264634056093052,
      "legacy_fitness": 0.5264634056093052,
      "target_cdf": 0.34803721968367785,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.1424522399902344,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5220211627705694,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.1424522399902344,
          "normalized": 0.34803721968367785,
          "contribution": 0.24362605377857446
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.415645718574524,
          "normalized": 0.9279836966399834,
          "contribution": 0.278395108991995
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/3b/3bb22c9025660aa6dae32e778579.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "fa075ab480b06d55cbf5b6730b419bc937227424108ae074fc49d0f84e5490dc",
      "id": "cc24aa0bc3c65eaf0d9b18c03cfe",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.332377910614014,
      "genuineness": 4.846253395080566,
      "wer": 0.07407407407407407,
      "identity": 0.7058227062225342,
      "voiceclap": -0.02812151238322258,
      "fitness": 0.5563795293408413,
      "legacy_fitness": 0.5563795293408413,
      "target_cdf": 0.4813064382554696,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.6919922828674316,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6096790487364303,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.6919922828674316,
          "normalized": 0.4813064382554696,
          "contribution": 0.33691450677882867
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.524202823638916,
          "normalized": 0.9092151398586724,
          "contribution": 0.2727645419576017
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/cc/cc24aa0bc3c65eaf0d9b18c03cfe.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "8f3822a66cc83478c55a2061f6e1392a9ade1f011a2c47cdc2d7e3e5e0f07d56",
      "id": "7d0852e65b74ef95e10d16b65533",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.332377910614014,
      "genuineness": 4.7458367347717285,
      "wer": 0.0,
      "identity": 0.5371086597442627,
      "voiceclap": -0.05062601715326309,
      "fitness": 0.5380772269732554,
      "legacy_fitness": 0.5380772269732554,
      "target_cdf": 0.36404916385149966,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.2111446857452393,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5142366278855512,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.2111446857452393,
          "normalized": 0.36404916385149966,
          "contribution": 0.2548344146960497
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7277473211288452,
          "normalized": 0.8646740439650051,
          "contribution": 0.2594022131895015
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/7d/7d0852e65b74ef95e10d16b65533.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "256d93ea9a7654cd3980406447454e43f9a011bdda4a41d0b71c7665ab2d848e",
      "id": "da429dc0f20611c77eb98920757c",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Wir sind schweigend an demselben Laden vorbeigegangen. Und irgendwie... habe ich es genau in diesem Moment verstanden. Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "sentences": [
        "Wir sind schweigend an demselben Laden vorbeigegangen.",
        "Und irgendwie... habe ich es genau in diesem Moment verstanden.",
        "Lass mich erzählen, was mir durch den Kopf gegangen ist."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.0 seconds duration] Wir sind schweigend an demselben Laden vorbeigegangen.\n(Natural clear speech.) [4.3 seconds duration] Und irgendwie... habe ich es genau in diesem Moment verstanden.\n(Natural clear speech.) [4.3 seconds duration] Lass mich erzählen, was mir durch den Kopf gegangen ist.",
      "requested_seconds": 11.6,
      "actual_seconds": 11.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 5.77639102935791,
      "genuineness": 4.444586753845215,
      "wer": 0.07407407407407407,
      "identity": 0.4269263744354248,
      "voiceclap": 0.03790433704853058,
      "fitness": 0.5114946026342694,
      "legacy_fitness": 0.5114946026342694,
      "target_cdf": 0.445626560735025,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.5483624935150146,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5321429245093283,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.5483624935150146,
          "normalized": 0.445626560735025,
          "contribution": 0.3119385925145175
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.1416213512420654,
          "normalized": 0.7340144399827025,
          "contribution": 0.22020433199481076
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/da/da429dc0f20611c77eb98920757c.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "a19d2ba4f3f10f1ecad4915b25d93e4c9450bed06436dcec0c938c22d5d4469b",
      "id": "fa230d4d9765e3e931c28e97ae7f",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 2,
      "generation": null,
      "ce": 5.601652145385742,
      "genuineness": 3.9425034523010254,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.05134008824825287,
      "fitness": 0.6342474781609118,
      "legacy_fitness": 0.6342474781609118,
      "target_cdf": 0.6598703946941175,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.425753116607666,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6888803624549467,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.425753116607666,
          "normalized": 0.6598703946941175,
          "contribution": 0.4619092762858822
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0805578231811523,
          "normalized": 0.7565702872302151,
          "contribution": 0.22697108616906453
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/fa/fa230d4d9765e3e931c28e97ae7f.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "eeae73a47500d521cfca52da643535e027be0250e8965065573c91a21b445009",
      "id": "38b3524674d6bc1c97bb12d931e8",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 2,
      "generation": null,
      "ce": 5.450741291046143,
      "genuineness": 4.055472373962402,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.0982772707939148,
      "fitness": 0.5958687639797322,
      "legacy_fitness": 0.5958687639797322,
      "target_cdf": 0.5812444322436238,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.0947799682617188,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6477025583695193,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.0947799682617188,
          "normalized": 0.5812444322436238,
          "contribution": 0.4068711025705366
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.9448615312576294,
          "normalized": 0.8027715193299423,
          "contribution": 0.24083145579898269
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/38/38b3524674d6bc1c97bb12d931e8.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "d1792b04650420751fd55b7a641094093397f1ca7ba8cd96c7cb7fb483ee42da",
      "id": "4b0c866e09f276d348472c545b1e",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 2,
      "generation": null,
      "ce": 6.2767791748046875,
      "genuineness": 4.331617832183838,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.08749891817569733,
      "fitness": 0.6637457755603453,
      "legacy_fitness": 0.6637457755603453,
      "target_cdf": 0.6526844396984696,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.394529104232788,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6871135558010165,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.394529104232788,
          "normalized": 0.6526844396984696,
          "contribution": 0.45687910778892865
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0500261783599854,
          "normalized": 0.7674481600402928,
          "contribution": 0.23023444801208784
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/4b/4b0c866e09f276d348472c545b1e.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "f11226e6e93e890aace5f1d557c6d0efc672911d67db4b5950a8359af7cbfb41",
      "id": "3eb9b51342078fd6aa939f642648",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 2,
      "generation": null,
      "ce": 6.078212261199951,
      "genuineness": 4.243753433227539,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.08533919602632523,
      "fitness": 0.7301641268638475,
      "legacy_fitness": 0.7301641268638475,
      "target_cdf": 0.8027715212320692,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 3.1282899379730225,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.8102651588094438,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 3.1282899379730225,
          "normalized": 0.8027715212320692,
          "contribution": 0.5619400648624484
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8634437322616577,
          "normalized": 0.8277503131566514,
          "contribution": 0.2483250939469954
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/3e/3eb9b51342078fd6aa939f642648.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "1f6c18a672757a90879750dfd6978d95cb39c307fba813b20f4d457c88d6d88a",
      "id": "065baab06a359f56d8d0887494c3",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 2,
      "generation": null,
      "ce": 6.348263263702393,
      "genuineness": 4.319066047668457,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.0668734461069107,
      "fitness": 0.6809441404750074,
      "legacy_fitness": 0.6809441404750074,
      "target_cdf": 0.684552947125857,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.535036563873291,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7090629609354688,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.535036563873291,
          "normalized": 0.684552947125857,
          "contribution": 0.4791870629880999
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0534186363220215,
          "normalized": 0.7662529931578967,
          "contribution": 0.229875897947369
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/06/065baab06a359f56d8d0887494c3.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "939392a3d24b727aee46f3afa43e1f9f2a394d3ba9fca4cdb4aa415b7886b850",
      "id": "2f8ac1db397d36caeb4952a4b3cd",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 2,
      "generation": null,
      "ce": 6.1894097328186035,
      "genuineness": 4.124508857727051,
      "wer": 0.0,
      "identity": 0.5275528430938721,
      "voiceclap": 0.07776005566120148,
      "fitness": 0.6653789307344917,
      "legacy_fitness": 0.6653789307344917,
      "target_cdf": 0.6775783033507989,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.503812789916992,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6850179731254432,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.503812789916992,
          "normalized": 0.6775783033507989,
          "contribution": 0.4743048123455592
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.223039150238037,
          "normalized": 0.7023772025996132,
          "contribution": 0.21071316077988397
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/2f/2f8ac1db397d36caeb4952a4b3cd.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "d5149e7cc9f51171fc1fb9bf0808c40feea6badfb75637818f8c592b7facddaf",
      "id": "aaa994938d22482a4167aeccc35d",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 2,
      "generation": null,
      "ce": 5.824047088623047,
      "genuineness": 4.042920112609863,
      "wer": 0.0,
      "identity": 0.5508952140808105,
      "voiceclap": 0.06119248643517494,
      "fitness": 0.6342275337384068,
      "legacy_fitness": 0.6342275337384068,
      "target_cdf": 0.6403427036615059,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.3414485454559326,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6813057053121193,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.3414485454559326,
          "normalized": 0.6403427036615059,
          "contribution": 0.44823989256305413
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0228869915008545,
          "normalized": 0.776886042496884,
          "contribution": 0.23306581274906518
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/aa/aaa994938d22482a4167aeccc35d.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "833853368d42319155930a3c012b329f5a6235c4d14eca5e7234582da4446f8f",
      "id": "13b7c301c4085778c6e83f667c15",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 2,
      "generation": null,
      "ce": 6.117925643920898,
      "genuineness": 3.898571252822876,
      "wer": 0.0,
      "identity": 0.7450971007347107,
      "voiceclap": 0.06257874518632889,
      "fitness": 0.6719447630255071,
      "legacy_fitness": 0.6719447630255071,
      "target_cdf": 0.7131123061197296,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.6661767959594727,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7026621898577468,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.6661767959594727,
          "normalized": 0.7131123061197296,
          "contribution": 0.4991786142838107
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.2824063301086426,
          "normalized": 0.6782785852464536,
          "contribution": 0.20348357557393607
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/13/13b7c301c4085778c6e83f667c15.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "6e58207321736bc3e52f0e88dbf6e79a657b0ab770cfddb8988b2fab8f412ac5",
      "id": "d26ef5368d7545d2e3e81b40dfca",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 2,
      "generation": null,
      "ce": 6.070269584655762,
      "genuineness": 4.243753433227539,
      "wer": 0.0,
      "identity": 0.6256559491157532,
      "voiceclap": 0.08198125660419464,
      "fitness": 0.6624339258804566,
      "legacy_fitness": 0.6624339258804566,
      "target_cdf": 0.6677082530924967,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.460099220275879,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6997599592361128,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.460099220275879,
          "normalized": 0.6677082530924967,
          "contribution": 0.46739577716474767
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0296716690063477,
          "normalized": 0.7745472735712169,
          "contribution": 0.23236418207136506
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d2/d26ef5368d7545d2e3e81b40dfca.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "1e186e277beb5d4f67e18b9326038531be265ba7cddb0c50a349e409dc91e5b0",
      "id": "d70b71f7a6df67e5c027f29f9678",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 2,
      "generation": null,
      "ce": 6.070269584655762,
      "genuineness": 4.218649387359619,
      "wer": 0.0,
      "identity": 0.6072022318840027,
      "voiceclap": 0.10131531953811646,
      "fitness": 0.6746375815332035,
      "legacy_fitness": 0.6746375815332035,
      "target_cdf": 0.6942075682203173,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.578749895095825,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7143768564914628,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.578749895095825,
          "normalized": 0.6942075682203173,
          "contribution": 0.48594529775422207
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.066988229751587,
          "normalized": 0.7614385291241359,
          "contribution": 0.22843155873724075
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d7/d70b71f7a6df67e5c027f29f9678.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "2b789f63469352eec54587e997bb0b3138eb6580e36ef90e523091b624c8c74d",
      "id": "76782700e77a00dd5ea11f5d7539",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 2,
      "generation": null,
      "ce": 5.700935363769531,
      "genuineness": 4.293961524963379,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.0590282566845417,
      "fitness": 0.5943736093951404,
      "legacy_fitness": 0.5943736093951404,
      "target_cdf": 0.5458703235215228,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.9511502981185913,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6235891755625194,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.9511502981185913,
          "normalized": 0.5458703235215228,
          "contribution": 0.3821092264650659
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.9380767345428467,
          "normalized": 0.8049331636581785,
          "contribution": 0.24147994909745352
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/76/76782700e77a00dd5ea11f5d7539.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "511afc42a580014b3c6fbf249d65b1ca0d51dade6e1694666e73c8380017f963",
      "id": "1a8abaffdddc5489b0c72b2db72d",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 2,
      "generation": null,
      "ce": 6.41180419921875,
      "genuineness": 4.507347106933594,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.04587865620851517,
      "fitness": 0.5582422783594051,
      "legacy_fitness": 0.5582422783594051,
      "target_cdf": 0.42028208784674,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.4453237056732178,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5575376558862912,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.4453237056732178,
          "normalized": 0.42028208784674,
          "contribution": 0.29419746149271797
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.673468828201294,
          "normalized": 0.8778006479785774,
          "contribution": 0.2633401943935732
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/1a/1a8abaffdddc5489b0c72b2db72d.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "5ba17b4990a2a80710ea60c65e124e7b6a3d6e788681edc6f436af8b6224ff7c",
      "id": "76f909a690da658244a1f9a6bd8e",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 2,
      "generation": null,
      "ce": 5.577824115753174,
      "genuineness": 4.256305694580078,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.03409509360790253,
      "fitness": 0.5718558237584334,
      "legacy_fitness": 0.5718558237584334,
      "target_cdf": 0.5101283005142014,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.8075205087661743,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6106186123812338,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.8075205087661743,
          "normalized": 0.5101283005142014,
          "contribution": 0.357089810359941
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8023803234100342,
          "normalized": 0.845096006737643,
          "contribution": 0.25352880202129285
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/76/76f909a690da658244a1f9a6bd8e.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "5531f15a0a3dc12691bca8381af9594d9b5f7b262d6490d4706c89ef86b46c37",
      "id": "93754a3d9adcbcd9e173a74b9eaf",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 2,
      "generation": null,
      "ce": 6.181467056274414,
      "genuineness": 4.532451152801514,
      "wer": 0.06896551724137931,
      "identity": null,
      "voiceclap": 0.03623568266630173,
      "fitness": 0.562978809170018,
      "legacy_fitness": 0.562978809170018,
      "target_cdf": 0.5225842708180436,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.8574787378311157,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6187773102348315,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.8574787378311157,
          "normalized": 0.5225842708180436,
          "contribution": 0.36580898957263047
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8091652393341064,
          "normalized": 0.8432277355406702,
          "contribution": 0.25296832066220104
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/93/93754a3d9adcbcd9e173a74b9eaf.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "fff124ae85f5b7052b745fb3d8ab3bba2d99d2b6395c0661ab5f489c7ede8b2b",
      "id": "d4c49e84b21ab3c3a90a92692747",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 2,
      "generation": null,
      "ce": 6.435632705688477,
      "genuineness": 4.632867813110352,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.06254026293754578,
      "fitness": 0.6418751527026034,
      "legacy_fitness": 0.6418751527026034,
      "target_cdf": 0.5758963523615869,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.072923421859741,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6722803428695695,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.072923421859741,
          "normalized": 0.5758963523615869,
          "contribution": 0.4031274466531108
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.58526611328125,
          "normalized": 0.8971763207215291,
          "contribution": 0.2691528962164587
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d4/d4c49e84b21ab3c3a90a92692747.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "e9a5c751d7ec3de0c609ad9fc149e342653f21cb1caf8967f500de8d1c6ccb16",
      "id": "c679bd3f79cb172bd3139c7d15fc",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 2,
      "generation": null,
      "ce": 5.935244083404541,
      "genuineness": 4.494794845581055,
      "wer": 0.0,
      "identity": 0.5892950296401978,
      "voiceclap": 0.08133667707443237,
      "fitness": 0.6239948385874869,
      "legacy_fitness": 0.6239948385874869,
      "target_cdf": 0.5766612358729922,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.0760457515716553,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6645745061528573,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.0760457515716553,
          "normalized": 0.5766612358729922,
          "contribution": 0.40366286511109456
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.707392930984497,
          "normalized": 0.8697054701392091,
          "contribution": 0.26091164104176273
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/c6/c679bd3f79cb172bd3139c7d15fc.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "42cc2564af77adea5a9a9c2ba1cbf2e058694a45b5d3c74d349d3672cb974d7f",
      "id": "dba7f36b62e83adbd00a85916e30",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.04,
      "cfg": 2,
      "generation": null,
      "ce": 6.086154937744141,
      "genuineness": 4.65797233581543,
      "wer": 0.0,
      "identity": 0.5504136085510254,
      "voiceclap": 0.034671980887651443,
      "fitness": 0.66421147217683,
      "legacy_fitness": 0.66421147217683,
      "target_cdf": 0.6359508361485003,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.322714328765869,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.688564950284589,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.322714328765869,
          "normalized": 0.6359508361485003,
          "contribution": 0.44516558530395023
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.917722225189209,
          "normalized": 0.811331216602129,
          "contribution": 0.2433993649806387
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/db/dba7f36b62e83adbd00a85916e30.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "1b68a5aa20537b4bffa44d68274fd0f21c6152b7db57cc6f63f9db8f5fd30921",
      "id": "8534e295cfaa8c51ab1af40eea05",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 2,
      "generation": null,
      "ce": 5.736677646636963,
      "genuineness": 4.3881025314331055,
      "wer": 0.0,
      "identity": 0.5853273868560791,
      "voiceclap": 0.0434977188706398,
      "fitness": 0.6420250382984918,
      "legacy_fitness": 0.6420250382984918,
      "target_cdf": 0.6315409833123765,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.3039798736572266,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7049418498460546,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.3039798736572266,
          "normalized": 0.6315409833123765,
          "contribution": 0.44207868831866354
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6802536249160767,
          "normalized": 0.8762105384246368,
          "contribution": 0.26286316152739103
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/85/8534e295cfaa8c51ab1af40eea05.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "b6cc946ede2b54725b7989ece97802792f778e7831475725ef4a824c4105d5a0",
      "id": "3fd490fa258aa7b3cb7e0c48a1b9",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 2,
      "generation": null,
      "ce": 6.022613525390625,
      "genuineness": 4.344170093536377,
      "wer": 0.0,
      "identity": 0.6533529162406921,
      "voiceclap": 0.05817766487598419,
      "fitness": 0.6206675213819451,
      "legacy_fitness": 0.6206675213819451,
      "target_cdf": 0.5781901920329943,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.0822906494140625,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6309679772823745,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.0822906494140625,
          "normalized": 0.5781901920329943,
          "contribution": 0.40473313442309594
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0873427391052246,
          "normalized": 0.7541161428642621,
          "contribution": 0.2262348428592786
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/3f/3fd490fa258aa7b3cb7e0c48a1b9.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "bbf4647caa36f474b728550c1ea73a51c19593b2f6b059e43323fefe79d1e41d",
      "id": "0a0a3783a175845c62d85e2ffe6c",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "en",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 2,
      "generation": null,
      "ce": 6.09409761428833,
      "genuineness": 4.2374773025512695,
      "wer": 0.0,
      "identity": 0.4702857732772827,
      "voiceclap": 0.09700172394514084,
      "fitness": 0.6391347695211724,
      "legacy_fitness": 0.6391347695211724,
      "target_cdf": 0.6204415497819893,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.2571442127227783,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7026141499267982,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.2571442127227783,
          "normalized": 0.6204415497819893,
          "contribution": 0.4343090848473925
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.598835825920105,
          "normalized": 0.894350216931352,
          "contribution": 0.2683050650794056
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/0a/0a0a3783a175845c62d85e2ffe6c.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "5385f95ae4429b5ce0a4316ac9db9b95130ba304377ff6aed9026e0afea7a5e1",
      "id": "d187f91c1a5f25f9678066845d51",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.71682071685791,
      "genuineness": 4.3881025314331055,
      "wer": 0.03125,
      "identity": null,
      "voiceclap": 0.04412943869829178,
      "fitness": 0.6334659617268331,
      "legacy_fitness": 0.6334659617268331,
      "target_cdf": 0.6562844483823239,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.4101409912109375,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6896335618797146,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.4101409912109375,
          "normalized": 0.6562844483823239,
          "contribution": 0.4593991138676267
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0500261783599854,
          "normalized": 0.7674481600402928,
          "contribution": 0.23023444801208784
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d1/d187f91c1a5f25f9678066845d51.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "39ed37596a02e6507d81e066bfce5db7ea626b37f31146e0d94461e54819b352",
      "id": "5a233fa3a1008efd2ea590da25c7",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.601652145385742,
      "genuineness": 4.293961524963379,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.06843435764312744,
      "fitness": 0.6243267122534806,
      "legacy_fitness": 0.6243267122534806,
      "target_cdf": 0.6107406901573925,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.216553211212158,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6931792880799866,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.216553211212158,
          "normalized": 0.6107406901573925,
          "contribution": 0.42751848311017476
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6395447254180908,
          "normalized": 0.8855360165660395,
          "contribution": 0.26566080496981187
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/5a/5a233fa3a1008efd2ea590da25c7.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "5596c029cf0b8ffd9f5d422c361de4bccdc03e39df1af59e89efb4dd0d6c8e33",
      "id": "c46bf0688383fa27d4b6de52d539",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.871703147888184,
      "genuineness": 4.287685871124268,
      "wer": 0.03125,
      "identity": null,
      "voiceclap": -0.008343121036887169,
      "fitness": 0.6085591968044106,
      "legacy_fitness": 0.6085591968044106,
      "target_cdf": 0.6054879639521904,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.1946964263916016,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6373757290433395,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.1946964263916016,
          "normalized": 0.6054879639521904,
          "contribution": 0.4238415747665333
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.1992921829223633,
          "normalized": 0.7117805142560208,
          "contribution": 0.21353415427680625
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/c4/c46bf0688383fa27d4b6de52d539.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "7264f7394452d5a4bf9f9e4da97f646051e4f840a6bc6a572f80a32f8ab2107e",
      "id": "0beef265633fa5ba65a4acd81f8f",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.68,
      "cfg": 2,
      "generation": null,
      "ce": 5.80021858215332,
      "genuineness": 4.381826400756836,
      "wer": 0.03125,
      "identity": null,
      "voiceclap": 0.017103608697652817,
      "fitness": 0.5700945563107012,
      "legacy_fitness": 0.5700945563107012,
      "target_cdf": 0.5218062774061958,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.854356288909912,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6276461942256186,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.854356288909912,
          "normalized": 0.5218062774061958,
          "contribution": 0.36526439418433704
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6870384216308594,
          "normalized": 0.8746060001376053,
          "contribution": 0.2623818000412816
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/0beef265633fa5ba65a4acd81f8f.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "28e57debba7ad30bfc45814d460658c7988db6900388be7c028d8ba0a64a5523",
      "id": "6dc5b74139b4446654871cf58287",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.68,
      "cfg": 2,
      "generation": null,
      "ce": 5.764476776123047,
      "genuineness": 4.300237655639648,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.08449633419513702,
      "fitness": 0.6256564874315214,
      "legacy_fitness": 0.6256564874315214,
      "target_cdf": 0.6047359980869197,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.1915740966796875,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6722361833914592,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.1915740966796875,
          "normalized": 0.6047359980869197,
          "contribution": 0.42331519866084377
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.856658935546875,
          "normalized": 0.829736615768718,
          "contribution": 0.2489209847306154
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/6d/6dc5b74139b4446654871cf58287.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "58114647dafeab29ff22e2bb1e732768d608f62ecb94b857bd0e471010fa8ae8",
      "id": "ed7cf06d91befe8b70d64e371ca5",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.498397350311279,
      "genuineness": 4.168440818786621,
      "wer": 0.03125,
      "identity": 0.5063897371292114,
      "voiceclap": 0.04031083360314369,
      "fitness": 0.5917857768775676,
      "legacy_fitness": 0.5917857768775676,
      "target_cdf": 0.599461345547841,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.169717311859131,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6829631362770618,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.169717311859131,
          "normalized": 0.599461345547841,
          "contribution": 0.4196229418834887
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.673468828201294,
          "normalized": 0.8778006479785774,
          "contribution": 0.2633401943935732
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/ed/ed7cf06d91befe8b70d64e371ca5.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "857c8f569e834bdd3f44006978495471dd53ab8bee9a5d5ae01a41de0ffdbc07",
      "id": "f38f66eb981a1767f6f502531394",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.959072113037109,
      "genuineness": 4.331617832183838,
      "wer": 0.0,
      "identity": 0.5941427946090698,
      "voiceclap": 0.01877606473863125,
      "fitness": 0.6521824948265862,
      "legacy_fitness": 0.6521824948265862,
      "target_cdf": 0.6454432313193303,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.3633053302764893,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7075366711766091,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.3633053302764893,
          "normalized": 0.6454432313193303,
          "contribution": 0.4518102619235312
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7752411365509033,
          "normalized": 0.8524213641769265,
          "contribution": 0.25572640925307794
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/f3/f38f66eb981a1767f6f502531394.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "943c082d7bf433f7f20cb13854f567e593df2c53e35edb67d9e3d6b778448253",
      "id": "fd3654e5da27606616b65cd7e9a7",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 2,
      "generation": null,
      "ce": 6.070269584655762,
      "genuineness": 4.507347106933594,
      "wer": 0.0,
      "identity": 0.560528039932251,
      "voiceclap": 0.01293763518333435,
      "fitness": 0.6097868872751022,
      "legacy_fitness": 0.6097868872751022,
      "target_cdf": 0.5404480364062836,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.9292936325073242,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6474665217008573,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.9292936325073242,
          "normalized": 0.5404480364062836,
          "contribution": 0.3783136254843985
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.58526611328125,
          "normalized": 0.8971763207215291,
          "contribution": 0.2691528962164587
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/fd/fd3654e5da27606616b65cd7e9a7.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "3bfee399ed6d134fa69bdec07a54fc1c75a54128898c5b4e5a1abe823e281317",
      "id": "00acafdf0e8501fe2e8beb0f2b11",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.752562999725342,
      "genuineness": 4.582659721374512,
      "wer": 0.0,
      "identity": 0.6154848337173462,
      "voiceclap": 0.040154505521059036,
      "fitness": 0.6341113964960086,
      "legacy_fitness": 0.6341113964960086,
      "target_cdf": 0.5987063328912076,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.166594982147217,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6842996886361978,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.166594982147217,
          "normalized": 0.5987063328912076,
          "contribution": 0.41909443302384525
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6463295221328735,
          "normalized": 0.8840175187078418,
          "contribution": 0.26520525561235253
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/00/00acafdf0e8501fe2e8beb0f2b11.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "0ef6832260430e95b4f5122d051824c1b6d8c804efc0825919ca3675d7708951",
      "id": "d4c515d490fc94c1c7d136564507",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 2,
      "generation": null,
      "ce": 5.728734970092773,
      "genuineness": 4.319066047668457,
      "wer": 0.0,
      "identity": 0.0529339574277401,
      "voiceclap": 0.051136989146471024,
      "fitness": 0.5945658827470963,
      "legacy_fitness": 0.5945658827470963,
      "target_cdf": 0.5427728463505159,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.938660740852356,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6648097020251047,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.938660740852356,
          "normalized": 0.5427728463505159,
          "contribution": 0.3799409924453611
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.2595947980880737,
          "normalized": 0.9495623652658118,
          "contribution": 0.28486870957974353
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d4/d4c515d490fc94c1c7d136564507.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "7ea8204f5c976ce0dad285ab31c939609c56623ee17b8d3f2fa4b7b407cd4c82",
      "id": "167ff5f9318ade6ec216c3ffb0d1",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 2,
      "generation": null,
      "ce": 5.458683967590332,
      "genuineness": 4.111956596374512,
      "wer": 0.03333333333333333,
      "identity": null,
      "voiceclap": 0.05547528341412544,
      "fitness": 0.5065976385979202,
      "legacy_fitness": 0.5065976385979202,
      "target_cdf": 0.4325357972976148,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.4952819347381592,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5468055014855879,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.4952819347381592,
          "normalized": 0.4325357972976148,
          "contribution": 0.30277505810833033
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.9109374284744263,
          "normalized": 0.8134348112575251,
          "contribution": 0.24403044337725754
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/16/167ff5f9318ade6ec216c3ffb0d1.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "a01e27197a3e7733cd6e223182ce7848c883e0d4f59c1feed3198eface6c69b6",
      "id": "d958bedc73a3c3ff281f29d702c7",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.88,
      "cfg": 2,
      "generation": null,
      "ce": 5.315715789794922,
      "genuineness": 4.168440818786621,
      "wer": 0.03333333333333333,
      "identity": null,
      "voiceclap": 0.05177262797951698,
      "fitness": 0.565424687384448,
      "legacy_fitness": 0.565424687384448,
      "target_cdf": 0.5566883230734452,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.994863748550415,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6562409648355695,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.994863748550415,
          "normalized": 0.5566883230734452,
          "contribution": 0.3896818261514116
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6259750127792358,
          "normalized": 0.8885304622805263,
          "contribution": 0.26655913868415787
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d9/d958bedc73a3c3ff281f29d702c7.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "856f2337bcbd03438f988b3948d2f58c6184f2bce397e7d178c8ceba86001ca7",
      "id": "a47b2ffab73233dad22d97fcf7b2",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 2,
      "generation": null,
      "ce": 5.33954381942749,
      "genuineness": 4.645420074462891,
      "wer": 0.03333333333333333,
      "identity": null,
      "voiceclap": 0.05655108392238617,
      "fitness": 0.5540470188386662,
      "legacy_fitness": 0.5540470188386662,
      "target_cdf": 0.4922086463976364,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.7357056140899658,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6044557854621022,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.7357056140899658,
          "normalized": 0.4922086463976364,
          "contribution": 0.34454605247834547
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7209625244140625,
          "normalized": 0.8663657766125228,
          "contribution": 0.2599097329837568
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/a4/a47b2ffab73233dad22d97fcf7b2.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "5982011320290110fdd5ff1e720822d719714daf4ee7f9a3ee9a5e1629506cea",
      "id": "b42287c31b22807b561650d4f51c",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 2,
      "generation": null,
      "ce": 5.839932441711426,
      "genuineness": 4.65797233581543,
      "wer": 0.1,
      "identity": null,
      "voiceclap": 0.0631384328007698,
      "fitness": 0.5293193110724711,
      "legacy_fitness": 0.5293193110724711,
      "target_cdf": 0.49610415231307836,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.7513176202774048,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6013577506457318,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.7513176202774048,
          "normalized": 0.49610415231307836,
          "contribution": 0.34727290661915483
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7955955266952515,
          "normalized": 0.84694948008859,
          "contribution": 0.254084844026577
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/b4/b42287c31b22807b561650d4f51c.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "7b878d5babedb1bc19c869eb391476db5ffbf9590b59e55ea4626807b4ff9784",
      "id": "b5bbc3ecf0f35debd958e2000f6a",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 2,
      "generation": null,
      "ce": 5.258131504058838,
      "genuineness": 4.143336772918701,
      "wer": 0.06666666666666667,
      "identity": null,
      "voiceclap": 0.04945123568177223,
      "fitness": 0.511699324624585,
      "legacy_fitness": 0.511699324624585,
      "target_cdf": 0.48831391315413425,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.7200937271118164,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6027313802496567,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.7200937271118164,
          "normalized": 0.48831391315413425,
          "contribution": 0.34181973920789394
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.707392930984497,
          "normalized": 0.8697054701392091,
          "contribution": 0.26091164104176273
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/b5/b5bbc3ecf0f35debd958e2000f6a.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "a05e0527f2c2fee9446015506c7784767e14775ce62ef6d2836dabddc38f0572",
      "id": "ced4149aa35a52ea367cea36463c",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 2,
      "generation": null,
      "ce": 5.609594821929932,
      "genuineness": 4.494794845581055,
      "wer": 0.03333333333333333,
      "identity": 0.46890509128570557,
      "voiceclap": 0.0341087281703949,
      "fitness": 0.5560126383277801,
      "legacy_fitness": 0.5560126383277801,
      "target_cdf": 0.4953249976223284,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.7481951713562012,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6300786901766628,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.7481951713562012,
          "normalized": 0.4953249976223284,
          "contribution": 0.3467274983356299
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.3003038167953491,
          "normalized": 0.9445039728034432,
          "contribution": 0.28335119184103297
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/ce/ced4149aa35a52ea367cea36463c.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "f15eb77acf40389679055821342b0889e248192ec4059f74d2dd15d06aa69ce8",
      "id": "0dc30cd9319315a6ab6e1cf13333",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 2,
      "generation": null,
      "ce": 5.808161735534668,
      "genuineness": 4.557555198669434,
      "wer": 0.03333333333333333,
      "identity": 0.4093482494354248,
      "voiceclap": 0.0432409830391407,
      "fitness": 0.516432073311432,
      "legacy_fitness": 0.516432073311432,
      "target_cdf": 0.3982757983463055,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.3547744750976562,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5366463685034564,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.3547744750976562,
          "normalized": 0.3982757983463055,
          "contribution": 0.27879305884241384
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7481017112731934,
          "normalized": 0.8595110322034752,
          "contribution": 0.2578533096610426
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/0dc30cd9319315a6ab6e1cf13333.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "51d0ecc0145216f592eaa62d4310036204666cfb7d83eecf071d7ab5c773dab7",
      "id": "c4d0c16fbbe3fcf2ec41aec8a1ff",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 2,
      "generation": null,
      "ce": 5.514282703399658,
      "genuineness": 4.243753433227539,
      "wer": 0.03333333333333333,
      "identity": 0.11758802831172943,
      "voiceclap": 0.04764532297849655,
      "fitness": 0.5251839139899627,
      "legacy_fitness": 0.5251839139899627,
      "target_cdf": 0.4572271536494841,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.5951982736587524,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5799687405383956,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.5951982736587524,
          "normalized": 0.4572271536494841,
          "contribution": 0.32005900755463884
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7209625244140625,
          "normalized": 0.8663657766125228,
          "contribution": 0.2599097329837568
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/c4/c4d0c16fbbe3fcf2ec41aec8a1ff.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "6431a32d1dc32961d2a86b97659f27113560052f99185dc4d877f603bf562758",
      "id": "524bc5fd3b290c46bbc506b68f83",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 2,
      "generation": null,
      "ce": 5.7128496170043945,
      "genuineness": 4.394378662109375,
      "wer": 0.0,
      "identity": 0.26270678639411926,
      "voiceclap": 0.06195581704378128,
      "fitness": 0.5619065860141433,
      "legacy_fitness": 0.5619065860141433,
      "target_cdf": 0.47197246933561904,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.6545236110687256,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6050110098347332,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.6545236110687256,
          "normalized": 0.47197246933561904,
          "contribution": 0.33038072853493333
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.490278720855713,
          "normalized": 0.9154342709993327,
          "contribution": 0.2746302812997998
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/52/524bc5fd3b290c46bbc506b68f83.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "62c7e7dff854b09b9bc8576879b47a8a032c2535f85847497a90ef5b38e7efd9",
      "id": "3a3e4ab92420d7f48759683ca37a",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 2,
      "generation": null,
      "ce": 5.871703147888184,
      "genuineness": 4.469690799713135,
      "wer": 0.06666666666666667,
      "identity": 0.2024538516998291,
      "voiceclap": -0.015310144051909447,
      "fitness": 0.47324508484750694,
      "legacy_fitness": 0.47324508484750694,
      "target_cdf": 0.34803721968367785,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.1424522399902344,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4913508398835922,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.1424522399902344,
          "normalized": 0.34803721968367785,
          "contribution": 0.24362605377857446
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8702285289764404,
          "normalized": 0.8257492870167258,
          "contribution": 0.24772478610501772
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/3a/3a3e4ab92420d7f48759683ca37a.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "181b649acf5b49d95725a331f22f144d7e611b319ce4fc2dab03c1837fe39a63",
      "id": "78154ac89bc163cde9b09ff3648c",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 2,
      "generation": null,
      "ce": 5.403085231781006,
      "genuineness": 4.331617832183838,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.05683975666761398,
      "fitness": 0.5877221679805806,
      "legacy_fitness": 0.5877221679805806,
      "target_cdf": 0.5443219216901246,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.9449055194854736,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6225052942805407,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.9449055194854736,
          "normalized": 0.5443219216901246,
          "contribution": 0.3810253451830872
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.9380767345428467,
          "normalized": 0.8049331636581785,
          "contribution": 0.24147994909745352
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/78/78154ac89bc163cde9b09ff3648c.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "90d60a5d2d2885f01913ccc46f5afbc090ea16cb18ca1fd6f424e507732c66f5",
      "id": "6dcd8b27d9b5058588dc899c1294",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 2,
      "generation": null,
      "ce": 5.148920059204102,
      "genuineness": 4.293961524963379,
      "wer": 0.02702702702702703,
      "identity": null,
      "voiceclap": 0.09255290776491165,
      "fitness": 0.6213848749710624,
      "legacy_fitness": 0.6213848749710624,
      "target_cdf": 0.662015001844475,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.435120105743408,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7290713062609444,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.435120105743408,
          "normalized": 0.662015001844475,
          "contribution": 0.4634105012911325
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6395447254180908,
          "normalized": 0.8855360165660395,
          "contribution": 0.26566080496981187
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/6d/6dcd8b27d9b5058588dc899c1294.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "106fd0d090cd97753dba58a088131350d8282a401a25d6536f8db706071db16b",
      "id": "1f40858c7efe6b41d333006f5cf6",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 2,
      "generation": null,
      "ce": 5.661221981048584,
      "genuineness": 4.457139015197754,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.07098601013422012,
      "fitness": 0.6528649602353681,
      "legacy_fitness": 0.6528649602353681,
      "target_cdf": 0.6512405701518276,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.38828444480896,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7126670856976351,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.38828444480896,
          "normalized": 0.6512405701518276,
          "contribution": 0.45586839910627924
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7616714239120483,
          "normalized": 0.8559956219711865,
          "contribution": 0.25679868659135596
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/1f/1f40858c7efe6b41d333006f5cf6.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "6c365228bfe1a5809e5a8d689ff09cb7fe9b4768cf5b09338ace663e33b2c1ab",
      "id": "d06d52ac65dd0bf6cc699cfbc620",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 2,
      "generation": null,
      "ce": 5.617537498474121,
      "genuineness": 4.356721878051758,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.027899619191884995,
      "fitness": 0.5700205918705488,
      "legacy_fitness": 0.5700205918705488,
      "target_cdf": 0.49610415231307836,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.7513176202774048,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5571724415612328,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.7513176202774048,
          "normalized": 0.49610415231307836,
          "contribution": 0.34727290661915483
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.2298238277435303,
          "normalized": 0.6996651164735933,
          "contribution": 0.209899534942078
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d0/d06d52ac65dd0bf6cc699cfbc620.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "28377faa566531f69b72e4a9c692071cc62bc99b06c9f01c68a98bbff32fb716",
      "id": "ceea058be378c330d6b7cbc2875a",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.92,
      "cfg": 2,
      "generation": null,
      "ce": 5.478540420532227,
      "genuineness": 4.362998008728027,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.030686169862747192,
      "fitness": 0.5936242010909457,
      "legacy_fitness": 0.5936242010909457,
      "target_cdf": 0.5497382137612778,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.9667621850967407,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6103113840750142,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.9667621850967407,
          "normalized": 0.5497382137612778,
          "contribution": 0.3848167496328945
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.094127655029297,
          "normalized": 0.7516487814737323,
          "contribution": 0.22549463444211967
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/ce/ceea058be378c330d6b7cbc2875a.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "d3879b0bcfaf2e2be2a9cdddfdf1f25467d62d4209c641e4bcacb1b89614a8d2",
      "id": "73f8e24961c7788d03e4904b4a90",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 2,
      "generation": null,
      "ce": 5.434855937957764,
      "genuineness": 4.3504462242126465,
      "wer": 0.0,
      "identity": 0.34521400928497314,
      "voiceclap": 0.02324468269944191,
      "fitness": 0.5850371482941679,
      "legacy_fitness": 0.5850371482941679,
      "target_cdf": 0.5357943143393938,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.9105592966079712,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6339463269277144,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.9105592966079712,
          "normalized": 0.5357943143393938,
          "contribution": 0.37505602003757565
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.734532117843628,
          "normalized": 0.8629676896337954,
          "contribution": 0.25889030689013864
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/73/73f8e24961c7788d03e4904b4a90.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "261b8dac48268bfec16585c003c84ba74a9833cc890c46f8d234f8b715f1f3be",
      "id": "848aca6138dd75db55b4a51f3ae1",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 2,
      "generation": null,
      "ce": 5.816103935241699,
      "genuineness": 4.582659721374512,
      "wer": 0.0,
      "identity": 0.4676440358161926,
      "voiceclap": 0.030217695981264114,
      "fitness": 0.6687799344257019,
      "legacy_fitness": 0.6687799344257019,
      "target_cdf": 0.6648663619747764,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.4476096630096436,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.730611708994696,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.4476096630096436,
          "normalized": 0.6648663619747764,
          "contribution": 0.4654064533823435
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6463295221328735,
          "normalized": 0.8840175187078418,
          "contribution": 0.26520525561235253
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/84/848aca6138dd75db55b4a51f3ae1.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "e57f056e43fe8b408d6353ce0de338a85531388b139e57ef358ea65bca0d534d",
      "id": "e92e1d713a1908331b3b28c4561e",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.76,
      "cfg": 2,
      "generation": null,
      "ce": 5.732706069946289,
      "genuineness": 4.482243061065674,
      "wer": 0.02702702702702703,
      "identity": 0.5370856523513794,
      "voiceclap": 0.009716052561998367,
      "fitness": 0.6229931878547561,
      "legacy_fitness": 0.6229931878547561,
      "target_cdf": 0.6204415497819893,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.2571442127227783,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6670245891605379,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.2571442127227783,
          "normalized": 0.6204415497819893,
          "contribution": 0.4343090848473925
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0262794494628906,
          "normalized": 0.7757183477104845,
          "contribution": 0.23271550431314536
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/e9/e92e1d713a1908331b3b28c4561e.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "ab8c58653fdfaa7601045cd4173b6ae8d47c27e975655b148f9986f4ca20d8b4",
      "id": "8312ae38a90c9308d6e2194b0b9b",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.76,
      "cfg": 2,
      "generation": null,
      "ce": 5.689021587371826,
      "genuineness": 4.532451152801514,
      "wer": 0.0,
      "identity": 0.5544670820236206,
      "voiceclap": 0.06024372577667236,
      "fitness": 0.6656404863233076,
      "legacy_fitness": 0.6656404863233076,
      "target_cdf": 0.6691256305445644,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.466344118118286,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.7444404558482782,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.466344118118286,
          "normalized": 0.6691256305445644,
          "contribution": 0.46838794138119505
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.4631394147872925,
          "normalized": 0.9201750482236108,
          "contribution": 0.27605251446708323
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/83/8312ae38a90c9308d6e2194b0b9b.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "38fcf2f4a3da6f17535b318aa655511c56c60787f88b52836321a2fbe9e1a743",
      "id": "bd38a1e861283efc972d71300439",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-selected",
      "language": "de",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 2,
      "generation": null,
      "ce": 5.736677646636963,
      "genuineness": 4.432034492492676,
      "wer": 0.0,
      "identity": 0.5829089879989624,
      "voiceclap": 0.03407283127307892,
      "fitness": 0.6247905227440919,
      "legacy_fitness": 0.6247905227440919,
      "target_cdf": 0.5934109554486127,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.14473819732666,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6819468074981867,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.14473819732666,
          "normalized": 0.5934109554486127,
          "contribution": 0.41538766881402883
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6259750127792358,
          "normalized": 0.8885304622805263,
          "contribution": 0.26655913868415787
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/bd/bd38a1e861283efc972d71300439.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "400bb86c687db5c40120b040206f57326cca14254372b4d647619caa778ecd2c",
      "id": "68d24201267d6693a76d505e2baf",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.55477237701416,
      "genuineness": 4.482243061065674,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.06824299693107605,
      "fitness": 0.6511145926829216,
      "legacy_fitness": 0.6511145926829216,
      "target_cdf": 0.6009703114263292,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.175962209701538,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6188549567020688,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.175962209701538,
          "normalized": 0.6009703114263292,
          "contribution": 0.4206792179984304
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.3248114585876465,
          "normalized": 0.6605857956787948,
          "contribution": 0.19817573870363844
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/68/68d24201267d6693a76d505e2baf.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "9f3a81e0d39f799c785d732da146a8489972382f524d4269fea3c21df359fbca",
      "id": "6d476befd27b9e1a7ef10db21a33",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.435632705688477,
      "genuineness": 4.494794845581055,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.0907282829284668,
      "fitness": 0.570601292705651,
      "legacy_fitness": 0.570601292705651,
      "target_cdf": 0.4448547129951234,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.545240044593811,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5551137552902686,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.545240044593811,
          "normalized": 0.4448547129951234,
          "contribution": 0.3113982990965864
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.9143297672271729,
          "normalized": 0.8123848539789408,
          "contribution": 0.24371545619368223
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/6d/6d476befd27b9e1a7ef10db21a33.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "14e44e3698cb260f8d5180d9e73671549841707370030048c52c674f2a782987",
      "id": "83fab06eecff5921bedd8e8f0a4a",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.372091293334961,
      "genuineness": 4.15588903427124,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.013881487771868706,
      "fitness": 0.5057615386837259,
      "legacy_fitness": 0.5057615386837259,
      "target_cdf": 0.34659442651143363,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.1362074613571167,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4938764223924207,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.1362074613571167,
          "normalized": 0.34659442651143363,
          "contribution": 0.24261609855800353
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8295196294784546,
          "normalized": 0.8375344127813906,
          "contribution": 0.25126032383441715
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/83/83fab06eecff5921bedd8e8f0a4a.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "11b2d4749b510b2704988e856149f0bc3e4be39726a376784c1d959416a66278",
      "id": "780d6973049978c2a4b3da26a670",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.443574905395508,
      "genuineness": 4.331617832183838,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.10003107786178589,
      "fitness": 0.4876896006291793,
      "legacy_fitness": 0.4876896006291793,
      "target_cdf": 0.2922323033065968,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.8926613330841064,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4358665456592721,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.8926613330841064,
          "normalized": 0.2922323033065968,
          "contribution": 0.20456261231461773
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.039849042892456,
          "normalized": 0.7710131111488479,
          "contribution": 0.23130393334465435
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/78/780d6973049978c2a4b3da26a670.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "af4851eb7f76565477d9ea664a0f8bd00fa290573ee73c7bdbb044e3be931c99",
      "id": "bc27a4d236ef51e4d0ff0ea411d2",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.586543083190918,
      "genuineness": 4.381826400756836,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.005364054813981056,
      "fitness": 0.5187382862572871,
      "legacy_fitness": 0.5187382862572871,
      "target_cdf": 0.3429972182919588,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.1205955743789673,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4989883596945098,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.1205955743789673,
          "normalized": 0.3429972182919588,
          "contribution": 0.24009805280437113
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.734532117843628,
          "normalized": 0.8629676896337954,
          "contribution": 0.25889030689013864
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/bc/bc27a4d236ef51e4d0ff0ea411d2.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "c5f8e59053144008cf6e4f3c2566c38da8b910aecb8a8a79c493b074b558c82d",
      "id": "5b6568b4ce3f139ce6428148a0cd",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.38003396987915,
      "genuineness": 4.268857479095459,
      "wer": 0.0,
      "identity": 0.7170006632804871,
      "voiceclap": 0.07818363606929779,
      "fitness": 0.5760777754669345,
      "legacy_fitness": 0.5760777754669345,
      "target_cdf": 0.47741572918195646,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.6763802766799927,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5816139821823545,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.6763802766799927,
          "normalized": 0.47741572918195646,
          "contribution": 0.3341910104273695
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8736209869384766,
          "normalized": 0.8247432391832834,
          "contribution": 0.247422971754985
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/5b/5b6568b4ce3f139ce6428148a0cd.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "ff780958ffd6d02625717f26997e63a2209a7ebe58bde8886c5b003262751769",
      "id": "4503b0e33256bcfabb475a20ecf4",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.419747352600098,
      "genuineness": 4.243753433227539,
      "wer": 0.0,
      "identity": 0.6285547614097595,
      "voiceclap": 0.01038043387234211,
      "fitness": 0.5768026282344653,
      "legacy_fitness": 0.5768026282344653,
      "target_cdf": 0.4789717694032973,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.6826250553131104,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5731400570355036,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.6826250553131104,
          "normalized": 0.4789717694032973,
          "contribution": 0.3352802385823081
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.975393295288086,
          "normalized": 0.7928660615106519,
          "contribution": 0.23785981845319554
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/45/4503b0e33256bcfabb475a20ecf4.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "2e80debbefa7b5261b5ec8fb0c1e0c229e96c881e69860203b91c36574aa8020",
      "id": "a6a063b88a7127ba2472a206e8a9",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.459460258483887,
      "genuineness": 4.287685871124268,
      "wer": 0.0,
      "identity": 0.6997462511062622,
      "voiceclap": 0.12663352489471436,
      "fitness": 0.5127170120101102,
      "legacy_fitness": 0.5127170120101102,
      "target_cdf": 0.3451538551690039,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.129962682723999,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.48532315481198496,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.129962682723999,
          "normalized": 0.3451538551690039,
          "contribution": 0.2416076986183027
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.9143297672271729,
          "normalized": 0.8123848539789408,
          "contribution": 0.24371545619368223
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/a6/a6a063b88a7127ba2472a206e8a9.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "05fa6a8fa07ab3bbe8dc0d529d9bd6bfcbc4147e18d9f1bde8b2a09e1ef60fa2",
      "id": "897f1d1cb09b7a0e5d88b1c58059",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.4912309646606445,
      "genuineness": 4.168440818786621,
      "wer": 0.0,
      "identity": 0.6847186088562012,
      "voiceclap": 0.031348779797554016,
      "fitness": 0.5583931647301871,
      "legacy_fitness": 0.5583931647301871,
      "target_cdf": 0.4448547129951234,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.545240044593811,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5779574377807443,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.545240044593811,
          "normalized": 0.4448547129951234,
          "contribution": 0.3113982990965864
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6259750127792358,
          "normalized": 0.8885304622805263,
          "contribution": 0.26655913868415787
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/89/897f1d1cb09b7a0e5d88b1c58059.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "ecb8762249b97c64a8f2e5f18f3bf2665305db203dd47e81147c5c7b6cb6a7fb",
      "id": "ff546e49fc79f76594dad2bebc45",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "I thought we would have more time before the train arrived. But here it is... and there is something I have not told you. Please stay with me until I have finished.",
      "sentences": [
        "I thought we would have more time before the train arrived.",
        "But here it is... and there is something I have not told you.",
        "Please stay with me until I have finished."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "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 we would have more time before the train arrived.\n(Natural clear speech.) [5.0 seconds duration] But here it is... and there is something I have not told you.\n(Natural clear speech.) [3.1 seconds duration] Please stay with me until I have finished.",
      "requested_seconds": 12.32,
      "actual_seconds": 12.32,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.316492557525635,
      "genuineness": 4.193545341491699,
      "wer": 0.0,
      "identity": 0.6389011144638062,
      "voiceclap": 0.0856119766831398,
      "fitness": 0.5365063723126638,
      "legacy_fitness": 0.5365063723126638,
      "target_cdf": 0.40772600495807115,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.3938043117523193,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.513476143296216,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.3938043117523193,
          "normalized": 0.40772600495807115,
          "contribution": 0.28540820347064977
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.070380687713623,
          "normalized": 0.7602264660852212,
          "contribution": 0.22806793982556633
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/ff/ff546e49fc79f76594dad2bebc45.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "b65a51c3eb58fd4b6486c4043f4bd2723d053a7f0456ad05ac4564c534968e93",
      "id": "02073a1354fd728c249eb4810b4d",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.403861999511719,
      "genuineness": 4.545003414154053,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.012964751571416855,
      "fitness": 0.542605083154752,
      "legacy_fitness": 0.542605083154752,
      "target_cdf": 0.38626678182108043,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.3048163652420044,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5095780508110843,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.3048163652420044,
          "normalized": 0.38626678182108043,
          "contribution": 0.27038674727475626
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.961823582649231,
          "normalized": 0.7973043451210936,
          "contribution": 0.23919130353632806
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/02/02073a1354fd728c249eb4810b4d.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "e0a0c4a64ee2f388515ef991b8dbc098cac047d0122af646497019623e247f01",
      "id": "532f9043a19c4552065e77909518",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.5786004066467285,
      "genuineness": 4.582659721374512,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.03823380917310715,
      "fitness": 0.4783510911475391,
      "legacy_fitness": 0.4783510911475391,
      "target_cdf": 0.24588385184819905,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.6678495407104492,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4511497554422638,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.6678495407104492,
          "normalized": 0.24588385184819905,
          "contribution": 0.17211869629373933
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.402076005935669,
          "normalized": 0.9301035304950818,
          "contribution": 0.2790310591485245
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/53/532f9043a19c4552065e77909518.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "3c24eeabc8ddf1198338068b16a98e955d40be3f3d3b8033cc6dbb7f09748e9e",
      "id": "d89f54498ddbb97ff9e03a793b90",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.546830177307129,
      "genuineness": 4.570107460021973,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.08309157192707062,
      "fitness": 0.49291574861577414,
      "legacy_fitness": 0.49291574861577414,
      "target_cdf": 0.2776477000310275,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.8239688277244568,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.42949909615769777,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.8239688277244568,
          "normalized": 0.2776477000310275,
          "contribution": 0.19435339002171925
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.002532482147217,
          "normalized": 0.7838190204532617,
          "contribution": 0.2351457061359785
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d8/d89f54498ddbb97ff9e03a793b90.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "a1d01175f17c171eb7097a6ac1c0c8b501507498457ea8328454b01fff8b5e0f",
      "id": "12769ea668f5f7d6360379f65a33",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.475345611572266,
      "genuineness": 4.494794845581055,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.013418247923254967,
      "fitness": 0.4403676469323791,
      "legacy_fitness": 0.4403676469323791,
      "target_cdf": 0.18240177615439046,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.318142294883728,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.3741921797987323,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.318142294883728,
          "normalized": 0.18240177615439046,
          "contribution": 0.12768124330807332
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8837981224060059,
          "normalized": 0.8217031216355299,
          "contribution": 0.24651093649065897
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/12/12769ea668f5f7d6360379f65a33.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "8820b6a3c45f88b367902ee9f396d35a1d01ef1c370e467f9da82eb5a2edd0cb",
      "id": "90dc53c4614815b02ee52446f7cc",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.562715530395508,
      "genuineness": 4.758388519287109,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.036570701748132706,
      "fitness": 0.49277566079754725,
      "legacy_fitness": 0.49277566079754725,
      "target_cdf": 0.26088316846806003,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.7427867650985718,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4132101943648486,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.7427867650985718,
          "normalized": 0.26088316846806003,
          "contribution": 0.18261821792764202
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.046633720397949,
          "normalized": 0.7686399214573554,
          "contribution": 0.2305919764372066
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/90/90dc53c4614815b02ee52446f7cc.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "a8be4c6b40f9c545f6c1094251227d70cae721512a3df045db9509514a3ee678",
      "id": "c071491abae1d3c444ed13b55d53",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.38003396987915,
      "genuineness": 4.482243061065674,
      "wer": 0.0,
      "identity": 0.6629244089126587,
      "voiceclap": 0.13382118940353394,
      "fitness": 0.49779006066058684,
      "legacy_fitness": 0.49779006066058684,
      "target_cdf": 0.30305816773841,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.9426195025444031,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4636884712752627,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.9426195025444031,
          "normalized": 0.30305816773841,
          "contribution": 0.21214071741688698
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.826127290725708,
          "normalized": 0.8384925128612525,
          "contribution": 0.25154775385837574
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/c0/c071491abae1d3c444ed13b55d53.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "1f1be380fcca78d0673ddd32ece849dc328df29cb4c57ea90180213205ad02dc",
      "id": "5dbf0be5be3ac37d54e72c238739",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.467403411865234,
      "genuineness": 4.381826400756836,
      "wer": 0.0,
      "identity": 0.4571056067943573,
      "voiceclap": 0.02210986614227295,
      "fitness": 0.4569875669124488,
      "legacy_fitness": 0.4569875669124488,
      "target_cdf": 0.22545276316856627,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.5616884231567383,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4067379189486118,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.5616884231567383,
          "normalized": 0.22545276316856627,
          "contribution": 0.1578169342179964
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.856658935546875,
          "normalized": 0.829736615768718,
          "contribution": 0.2489209847306154
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/5d/5dbf0be5be3ac37d54e72c238739.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "bd013105dd697bbd7ba0b71ae0fd5e3a25be2ba67e8c7c398e4fe29704dac175",
      "id": "16445224ca0d683acd4225ffd04e",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.4515180587768555,
      "genuineness": 4.645420074462891,
      "wer": 0.0,
      "identity": 0.6566537022590637,
      "voiceclap": 0.10301821678876877,
      "fitness": 0.5368717031644583,
      "legacy_fitness": 0.5368717031644583,
      "target_cdf": 0.36404916385149966,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.2111446857452393,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5040516924704357,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.2111446857452393,
          "normalized": 0.36404916385149966,
          "contribution": 0.2548344146960497
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8532664775848389,
          "normalized": 0.8307242592479531,
          "contribution": 0.24921727777438593
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/16/16445224ca0d683acd4225ffd04e.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "d2293916429a5ac203571c2fc286839afafb155b80a6d05c1a40da56edba8850",
      "id": "9409cd618414440396bb1c7aa337",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.475345611572266,
      "genuineness": 4.582659721374512,
      "wer": 0.0,
      "identity": 0.5287508964538574,
      "voiceclap": 0.035255856812000275,
      "fitness": 0.5268464051884169,
      "legacy_fitness": 0.5268464051884169,
      "target_cdf": 0.34803721968367785,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.1424522399902344,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.49104902553355945,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.1424522399902344,
          "normalized": 0.34803721968367785,
          "contribution": 0.24362605377857446
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8736209869384766,
          "normalized": 0.8247432391832834,
          "contribution": 0.247422971754985
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/94/9409cd618414440396bb1c7aa337.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "fce54d2eee0b65f4f486b6b7363cfd8dbab08495f8ee0d6ef069cdba66c50120",
      "id": "f9ea68eba1e6e8df8c58476c77f7",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "en",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "We walked past the same shop without saying a word. And somehow... that was the moment I finally understood. Let me tell you what was going through my mind.",
      "sentences": [
        "We walked past the same shop without saying a word.",
        "And somehow... that was the moment I finally understood.",
        "Let me tell you what was going through my mind."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] We walked past the same shop without saying a word.\n(Natural clear speech.) [3.5 seconds duration] And somehow... that was the moment I finally understood.\n(Natural clear speech.) [3.8 seconds duration] Let me tell you what was going through my mind.",
      "requested_seconds": 11.12,
      "actual_seconds": 11.12,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.372091293334961,
      "genuineness": 4.494794845581055,
      "wer": 0.0,
      "identity": 0.6070485711097717,
      "voiceclap": 0.09758737683296204,
      "fitness": 0.5561539429262558,
      "legacy_fitness": 0.5561539429262558,
      "target_cdf": 0.41913708405400896,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.4406402111053467,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5358405142028276,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.4406402111053467,
          "normalized": 0.41913708405400896,
          "contribution": 0.2933959588378062
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.9278994798660278,
          "normalized": 0.8081485178834047,
          "contribution": 0.2424445553650214
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/f9/f9ea68eba1e6e8df8c58476c77f7.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "18cd5d471f069bde2b4af17dfda6f6445feb98f1881d9529626275a75f0c074f",
      "id": "0829c846d5c1c75e7be4cf92f56d",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.483288764953613,
      "genuineness": 4.758388519287109,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.006778644397854805,
      "fitness": 0.5619398393365863,
      "legacy_fitness": 0.5619398393365863,
      "target_cdf": 0.4031828638182329,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.3750700950622559,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5505330697521686,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.3750700950622559,
          "normalized": 0.4031828638182329,
          "contribution": 0.282228004672763
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.598835825920105,
          "normalized": 0.894350216931352,
          "contribution": 0.2683050650794056
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/08/0829c846d5c1c75e7be4cf92f56d.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "ea995ecace7cf5c2afd4fd29d82acd3b0580de57c450ca72e4b92bef99693199",
      "id": "798c984d07b197e4996c0e507487",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.157639026641846,
      "genuineness": 4.632867813110352,
      "wer": 0.03125,
      "identity": null,
      "voiceclap": 0.031381987035274506,
      "fitness": 0.5639928974578086,
      "legacy_fitness": 0.5639928974578086,
      "target_cdf": 0.47041816436999373,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.648278832435608,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5902043561007584,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.648278832435608,
          "normalized": 0.47041816436999373,
          "contribution": 0.3292927150589956
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.707392930984497,
          "normalized": 0.8697054701392091,
          "contribution": 0.26091164104176273
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/79/798c984d07b197e4996c0e507487.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "8c5154c025ab93a2506529c29655868a0b44e88d4fc144ce86bec3c6d295b66d",
      "id": "b8f423b9381a087497d4c597a723",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.41180419921875,
      "genuineness": 4.281409740447998,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.07411050796508789,
      "fitness": 0.4531806167187235,
      "legacy_fitness": 0.4531806167187235,
      "target_cdf": 0.2289868784391763,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.5804227590560913,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4390057260717969,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.5804227590560913,
          "normalized": 0.2289868784391763,
          "contribution": 0.16029081490742342
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.4088608026504517,
          "normalized": 0.9290497038812449,
          "contribution": 0.27871491116437347
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/b8/b8f423b9381a087497d4c597a723.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "e4b5040dc641aa3b7e68dc16546d00b20d576ce2b1e0955015fa34fcee03872b",
      "id": "23a48c9e3797295ab1f5b7aa5a3e",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.3641486167907715,
      "genuineness": 4.532451152801514,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.020469002425670624,
      "fitness": 0.49676723518063604,
      "legacy_fitness": 0.49676723518063604,
      "target_cdf": 0.29762277678827403,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.9176404476165771,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4748950824359497,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.9176404476165771,
          "normalized": 0.29762277678827403,
          "contribution": 0.20833594375179182
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6259750127792358,
          "normalized": 0.8885304622805263,
          "contribution": 0.26655913868415787
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/23/23a48c9e3797295ab1f5b7aa5a3e.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "81d02f52a5d66f119bfb4ca9fc2b7b5cd211e1203cb04ad2f3531686c002ab28",
      "id": "c2846e6496dad16434f211c83042",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.2767791748046875,
      "genuineness": 4.570107460021973,
      "wer": 0.03125,
      "identity": null,
      "voiceclap": -0.03293399512767792,
      "fitness": 0.5963874571831853,
      "legacy_fitness": 0.5963874571831853,
      "target_cdf": 0.5365702774318226,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.9136816263198853,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5977145355628664,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.9136816263198853,
          "normalized": 0.5365702774318226,
          "contribution": 0.37559919420227583
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.124659299850464,
          "normalized": 0.7403844712019686,
          "contribution": 0.22211534136059058
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/c2/c2846e6496dad16434f211c83042.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "66aa4d7cda9fcf6fb5ebe3d80fef60a476776b838272190150e6ae7378e1031e",
      "id": "1181037282b8f1c7784cecc528f4",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.427689552307129,
      "genuineness": 4.494794845581055,
      "wer": 0.03125,
      "identity": 0.4917713403701782,
      "voiceclap": -0.06966041028499603,
      "fitness": 0.49805494236395376,
      "legacy_fitness": 0.49805494236395376,
      "target_cdf": 0.3322917469075033,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.0737597942352295,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.46139839206587746,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.0737597942352295,
          "normalized": 0.3322917469075033,
          "contribution": 0.23260422283525228
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.063595771789551,
          "normalized": 0.7626472307687506,
          "contribution": 0.22879416923062518
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/11/1181037282b8f1c7784cecc528f4.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "a02b921fc06ffd2886b07767a857b49b1423589a94f9b0c9f27cc8d6f3b98acf",
      "id": "e2deca4739e6cc713baeee83a22c",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.197352409362793,
      "genuineness": 4.65797233581543,
      "wer": 0.0,
      "identity": 0.6348891258239746,
      "voiceclap": -0.030305340886116028,
      "fitness": 0.6361988267500442,
      "legacy_fitness": 0.6361988267500442,
      "target_cdf": 0.5743656717139966,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 2.066678524017334,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6624688521524413,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 2.066678524017334,
          "normalized": 0.5743656717139966,
          "contribution": 0.40205597019979755
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7141776084899902,
          "normalized": 0.8680429398421459,
          "contribution": 0.2604128819526437
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/e2/e2deca4739e6cc713baeee83a22c.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "52607db2ad864ea88439c03e80a6149e70f68f979fe2479f8365bd9cc6a903c0",
      "id": "4e659e31484884b6c679504588f5",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.395919322967529,
      "genuineness": 4.582659721374512,
      "wer": 0.0,
      "identity": 0.45950883626937866,
      "voiceclap": -0.09939403831958771,
      "fitness": 0.4483755508385968,
      "legacy_fitness": 0.4483755508385968,
      "target_cdf": 0.19506682541427445,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.3930795192718506,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.3635178639590566,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.3930795192718506,
          "normalized": 0.19506682541427445,
          "contribution": 0.1365467777899921
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0805578231811523,
          "normalized": 0.7565702872302151,
          "contribution": 0.22697108616906453
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/4e/4e659e31484884b6c679504588f5.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "ef53ab796ec44f598619fb6a9d9ac06a702434eb5b758f9a625e9d0b89101b7c",
      "id": "7e02ae27b2970bbd061fb3eaf7c8",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.403861999511719,
      "genuineness": 4.482243061065674,
      "wer": 0.0,
      "identity": 0.6386200189590454,
      "voiceclap": -0.044415056705474854,
      "fitness": 0.4560956737733872,
      "legacy_fitness": 0.4560956737733872,
      "target_cdf": 0.2184779924823823,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.5242197513580322,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4271996125508377,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.5242197513580322,
          "normalized": 0.2184779924823823,
          "contribution": 0.1529345947376676
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.4970635175704956,
          "normalized": 0.9142167260439005,
          "contribution": 0.27426501781317014
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/7e/7e02ae27b2970bbd061fb3eaf7c8.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "9c3bcece1e67323084d421013a86f74ca0a0f5ffe75ab9abe890edb79328aeb4",
      "id": "f1b3c30d5fa41c856033c14d97e4",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen. Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern. Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "sentences": [
        "Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.",
        "Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.",
        "Lass mich darauf antworten, bevor wir diesen Ort verlassen."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Das letzte Licht im Gebäude ist vor einer Minute ausgegangen.\n(Natural clear speech.) [5.5 seconds duration] Ich war noch hier... und habe versucht, mich an deine Frage zu erinnern.\n(Natural clear speech.) [3.8 seconds duration] Lass mich darauf antworten, bevor wir diesen Ort verlassen.",
      "requested_seconds": 13.6,
      "actual_seconds": 13.6,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.427689552307129,
      "genuineness": 4.758388519287109,
      "wer": 0.0,
      "identity": 0.5424050092697144,
      "voiceclap": -0.037664733827114105,
      "fitness": 0.5417170109049098,
      "legacy_fitness": 0.5417170109049098,
      "target_cdf": 0.365517167587204,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.2173895835876465,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5142360158870121,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.2173895835876465,
          "normalized": 0.365517167587204,
          "contribution": 0.2558620173110428
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7413170337677002,
          "normalized": 0.8612466619198977,
          "contribution": 0.2583739985759693
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/f1/f1b3c30d5fa41c856033c14d97e4.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "5c9da5ed24ce22813518241736dfbc32ccc19dc1ddac5f46cd5787e6abc5c97b",
      "id": "b937e4c6a72cad73fb6e9e393e72",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 1.0,
      "generation": null,
      "ce": 5.502368927001953,
      "genuineness": 4.218649387359619,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.07799440622329712,
      "fitness": 0.5578826663080149,
      "legacy_fitness": 0.5578826663080149,
      "target_cdf": 0.4890927706526305,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.7232160568237305,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6204365809625481,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.7232160568237305,
          "normalized": 0.4890927706526305,
          "contribution": 0.3423649394568413
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.4224305152893066,
          "normalized": 0.9269054716856889,
          "contribution": 0.2780716415057067
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/b9/b937e4c6a72cad73fb6e9e393e72.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "e1442fefce2fbeb8d7603539f2be00d63a387a83d9571a43f171463aeb7ebdb6",
      "id": "33abb32e20411ff9465081563a8f",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.221180438995361,
      "genuineness": 4.507347106933594,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.03377341851592064,
      "fitness": 0.44746184175250836,
      "legacy_fitness": 0.44746184175250836,
      "target_cdf": 0.20825240264411576,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.4680168628692627,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.42720832518445595,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.4680168628692627,
          "normalized": 0.20825240264411576,
          "contribution": 0.14577668185088102
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.3477975130081177,
          "normalized": 0.938105477778583,
          "contribution": 0.2814316433335749
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/33/33abb32e20411ff9465081563a8f.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "f785083f6c001255774944701c0a369ec7c1039bb92d411d8875cbce14a23973",
      "id": "f3bbf1b5bd428ae096dddd290a62",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.340320587158203,
      "genuineness": 4.570107460021973,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.044916633516550064,
      "fitness": 0.44064773446114724,
      "legacy_fitness": 0.44064773446114724,
      "target_cdf": 0.18343715122921994,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.3243870735168457,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.3917462002540272,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.3243870735168457,
          "normalized": 0.18343715122921994,
          "contribution": 0.12840600586045395
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.673468828201294,
          "normalized": 0.8778006479785774,
          "contribution": 0.2633401943935732
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/f3/f3bbf1b5bd428ae096dddd290a62.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "1110335650e0c8ea95dadf50ad2643c30c0a942cba5825d23c99a2fb14becda9",
      "id": "e8acbb364984319f195f5d45fb19",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.284721851348877,
      "genuineness": 4.7081804275512695,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.00897701270878315,
      "fitness": 0.5494214552851455,
      "legacy_fitness": 0.5494214552851455,
      "target_cdf": 0.39225844904024154,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.3297953605651855,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5428859794075747,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.3297953605651855,
          "normalized": 0.39225844904024154,
          "contribution": 0.27458091432816906
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.598835825920105,
          "normalized": 0.894350216931352,
          "contribution": 0.2683050650794056
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/e8/e8acbb364984319f195f5d45fb19.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "69b3306f6f3b4255fd6709cc3e20beb7d6218e163ca4bdd8572c39830078cd20",
      "id": "fccda83abee9409e579d78e62d5c",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.88,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.41180419921875,
      "genuineness": 4.607763767242432,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.003934606909751892,
      "fitness": 0.4466196535554615,
      "legacy_fitness": 0.4466196535554615,
      "target_cdf": 0.18866878321311614,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.35561084747314453,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.40741727455774773,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.35561084747314453,
          "normalized": 0.18866878321311614,
          "contribution": 0.13206814824918128
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.476709008216858,
          "normalized": 0.9178304210285549,
          "contribution": 0.27534912630856645
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/fc/fccda83abee9409e579d78e62d5c.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "2ea6c025f4a1271617a2e85a3052c98f84def62012ecc5e02c97d664f21c9d2e",
      "id": "317b8bf95c04483fbf0416f5821e",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.38797664642334,
      "genuineness": 4.65797233581543,
      "wer": 0.0,
      "identity": 0.6072532534599304,
      "voiceclap": -0.036103926599025726,
      "fitness": 0.4945765290292329,
      "legacy_fitness": 0.4945765290292329,
      "target_cdf": 0.2815898644193463,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.8427031636238098,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4427018893329181,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.8427031636238098,
          "normalized": 0.2815898644193463,
          "contribution": 0.1971129050935424
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8939753770828247,
          "normalized": 0.8186299474645856,
          "contribution": 0.24558898423937567
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/31/317b8bf95c04483fbf0416f5821e.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "31d85b40641aef56e961461fda68fdee2bd4d7904a0b341a267a63092eca4f43",
      "id": "d0e00ecc4df96c8ffcc25f7afab2",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.459460258483887,
      "genuineness": 4.607763767242432,
      "wer": 0.03333333333333333,
      "identity": 0.48994380235671997,
      "voiceclap": -0.052574142813682556,
      "fitness": 0.4495325138977197,
      "legacy_fitness": 0.4495325138977197,
      "target_cdf": 0.22311394327214934,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.5491988658905029,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4293253766677677,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.5491988658905029,
          "normalized": 0.22311394327214934,
          "contribution": 0.15617976029050454
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.5174179077148438,
          "normalized": 0.9104853879242107,
          "contribution": 0.2731456163772632
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/d0/d0e00ecc4df96c8ffcc25f7afab2.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "2722f26d17984dbfb5c90f6f5cb5dd5b5febebf9d864381f9cb2c29925817c02",
      "id": "4490e5f7c581149bab3956c96005",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.443574905395508,
      "genuineness": 4.796044826507568,
      "wer": 0.03333333333333333,
      "identity": 0.6160815954208374,
      "voiceclap": -0.06854478269815445,
      "fitness": 0.4940556335720738,
      "legacy_fitness": 0.4940556335720738,
      "target_cdf": 0.30033492198072365,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.9301300048828125,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.46541810047684734,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.9301300048828125,
          "normalized": 0.30033492198072365,
          "contribution": 0.21023444538650654
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7820258140563965,
          "normalized": 0.8506121836344694,
          "contribution": 0.2551836550903408
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/44/4490e5f7c581149bab3956c96005.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "fba9ca66f1921c8d6d3dc57ce4ef3f173e2c77947fb02deee380fbc6a1770ac9",
      "id": "63208c007ca23a33ca4e0e235cce",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.24500846862793,
      "genuineness": 4.457139015197754,
      "wer": 0.0,
      "identity": 0.5653250217437744,
      "voiceclap": -0.00621374137699604,
      "fitness": 0.485949091991951,
      "legacy_fitness": 0.485949091991951,
      "target_cdf": 0.2882195092860261,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.8739270567893982,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.48402812098355996,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.8739270567893982,
          "normalized": 0.2882195092860261,
          "contribution": 0.20175365650021823
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.32744300365448,
          "normalized": 0.9409148816111391,
          "contribution": 0.2822744644833417
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/63/63208c007ca23a33ca4e0e235cce.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "aa082bf24bcfba913b17b27270b201dd6f044bb629d18424314e608ffbe87191",
      "id": "eed450ea454c9820bee67d7da7c1",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Jemand hat deinen Schirm neben das schmale Gartentor gestellt. Ich habe ihn reingeholt... und dann gesehen, was darunter lag. Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "sentences": [
        "Jemand hat deinen Schirm neben das schmale Gartentor gestellt.",
        "Ich habe ihn reingeholt... und dann gesehen, was darunter lag.",
        "Sieh dir das lieber an, bevor ich es zu erklären versuche."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Jemand hat deinen Schirm neben das schmale Gartentor gestellt.\n(Natural clear speech.) [4.3 seconds duration] Ich habe ihn reingeholt... und dann gesehen, was darunter lag.\n(Natural clear speech.) [4.7 seconds duration] Sieh dir das lieber an, bevor ich es zu erklären versuche.",
      "requested_seconds": 12.8,
      "actual_seconds": 12.8,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.173524379730225,
      "genuineness": 4.570107460021973,
      "wer": 0.06666666666666667,
      "identity": 0.629022479057312,
      "voiceclap": 0.025187049061059952,
      "fitness": 0.47750732407405644,
      "legacy_fitness": 0.47750732407405644,
      "target_cdf": 0.33371147283701674,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.0800044536590576,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5063625729435134,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.0800044536590576,
          "normalized": 0.33371147283701674,
          "contribution": 0.2335980309859117
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.524202823638916,
          "normalized": 0.9092151398586724,
          "contribution": 0.2727645419576017
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/ee/eed450ea454c9820bee67d7da7c1.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "68fe1ba878694e08cb5ccbcb62fd5ce2f15ee155fc1953d0413ee2432975eb6e",
      "id": "1808b1867397ae0a38e61949cdcb",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.284721851348877,
      "genuineness": 4.7332844734191895,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.022330954670906067,
      "fitness": 0.5272751567769403,
      "legacy_fitness": 0.5272751567769403,
      "target_cdf": 0.34587384820150413,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.133085012435913,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4741234926782728,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.133085012435913,
          "normalized": 0.34587384820150413,
          "contribution": 0.24211169374105287
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.033064126968384,
          "normalized": 0.7733726631240666,
          "contribution": 0.23201179893721996
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/18/1808b1867397ae0a38e61949cdcb.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "f7e0fc9233aa5d26d857518c22593f1b47acac1bf3ae99b063af229dc80c5780",
      "id": "0e84eff0e089629809607ca7d4d9",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.332377910614014,
      "genuineness": 4.796044826507568,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.04061030596494675,
      "fitness": 0.5339715422907924,
      "legacy_fitness": 0.5339715422907924,
      "target_cdf": 0.35165378684192,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.1580641269683838,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.49448274473633935,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.1580641269683838,
          "normalized": 0.35165378684192,
          "contribution": 0.24615765078934398
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8634437322616577,
          "normalized": 0.8277503131566514,
          "contribution": 0.2483250939469954
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/0e84eff0e089629809607ca7d4d9.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "d61f7f10b823393668717f77c2b85989392cebe81abd6b64e41b92a22a34aa1f",
      "id": "09ba636657cc07fca0a1b6f645c7",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.070269584655762,
      "genuineness": 4.65797233581543,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.0045091453939676285,
      "fitness": 0.6133480129957869,
      "legacy_fitness": 0.6133480129957869,
      "target_cdf": 0.5350181854408332,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.9074368476867676,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6318409279317989,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.9074368476867676,
          "normalized": 0.5350181854408332,
          "contribution": 0.37451272980858324
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7548866271972656,
          "normalized": 0.8577606604107191,
          "contribution": 0.2573281981232157
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/09/09ba636657cc07fca0a1b6f645c7.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "52e98b8ebbe72d425e9007c5912794dc7c72b48be30189e39a6696881e5781b2",
      "id": "a50622c90a01154e9ce6ae9f57a8",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.92,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.141753673553467,
      "genuineness": 4.65797233581543,
      "wer": 0.0,
      "identity": null,
      "voiceclap": 0.030292611569166183,
      "fitness": 0.5766269833958918,
      "legacy_fitness": 0.5766269833958918,
      "target_cdf": 0.4580019217961578,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.5983206033706665,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5862621502271224,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.5983206033706665,
          "normalized": 0.4580019217961578,
          "contribution": 0.32060134525731043
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.6395447254180908,
          "normalized": 0.8855360165660395,
          "contribution": 0.26566080496981187
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/a5/a50622c90a01154e9ce6ae9f57a8.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "f679c94f7feea77886ff7d540245c525d3b4a77a425a26e872fbda7ecc78633a",
      "id": "263c1d8ec4d3c46d62ab0d0c9d41",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": false,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.300607204437256,
      "genuineness": 4.68307638168335,
      "wer": 0.0,
      "identity": null,
      "voiceclap": -0.0005227560177445412,
      "fitness": 0.5940749804214647,
      "legacy_fitness": 0.5940749804214647,
      "target_cdf": 0.4828632354807873,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.6982370615005493,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.5968945717266898,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.6982370615005493,
          "normalized": 0.4828632354807873,
          "contribution": 0.3380042648365511
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.734532117843628,
          "normalized": 0.8629676896337954,
          "contribution": 0.25889030689013864
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/26/263c1d8ec4d3c46d62ab0d0c9d41.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "668c78e52e84b8276911c387df7b233b63d9db33f1fc9cd45af2c08344dc0e3f",
      "id": "e2286bf35a14da8cdc297e9789b2",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1701,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.340320587158203,
      "genuineness": 4.68307638168335,
      "wer": 0.0,
      "identity": 0.46323320269584656,
      "voiceclap": -0.048474445939064026,
      "fitness": 0.46694987624443735,
      "legacy_fitness": 0.46694987624443735,
      "target_cdf": 0.22662735799068534,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.567933201789856,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.40484386583482423,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.567933201789856,
          "normalized": 0.22662735799068534,
          "contribution": 0.15863915059347972
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.887190580368042,
          "normalized": 0.820682384137815,
          "contribution": 0.2462047152413445
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/e2/e2286bf35a14da8cdc297e9789b2.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "da52f6fc9fa704d4f1b01a48f2f79950b9db3a469444f91078700e7f31c70bc8",
      "id": "96f0412ac1f389ca9590ed08d449",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1702,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.284721851348877,
      "genuineness": 4.570107460021973,
      "wer": 0.0,
      "identity": 0.6727700233459473,
      "voiceclap": -0.017276104539632797,
      "fitness": 0.5186789164412613,
      "legacy_fitness": 0.5186789164412613,
      "target_cdf": 0.3422794519799145,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.1174731254577637,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4942320742885176,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.1174731254577637,
          "normalized": 0.3422794519799145,
          "contribution": 0.23959561638594012
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.7888107299804688,
          "normalized": 0.8487881930085917,
          "contribution": 0.2546364579025775
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/96/96f0412ac1f389ca9590ed08d449.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "dde089db3933ccb318d19d58a6e9a7dade4881404cbb9f974512b8e534ee68ac",
      "id": "4362fb80a7a196ae06e304fbd6d9",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1703,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.38003396987915,
      "genuineness": 4.6705241203308105,
      "wer": 0.0,
      "identity": 0.5038354396820068,
      "voiceclap": -0.07954525202512741,
      "fitness": 0.519543205655147,
      "legacy_fitness": 0.519543205655147,
      "target_cdf": 0.3308743694554356,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.0675148963928223,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.4650271390093567,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.0675148963928223,
          "normalized": 0.3308743694554356,
          "contribution": 0.23161205861880488
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 2.0194945335388184,
          "normalized": 0.7780502679685062,
          "contribution": 0.23341508039055184
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/43/4362fb80a7a196ae06e304fbd6d9.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "b7a95507ad01cb884751d15ed9cbde53227bb087bd9a8dd0b2188d86f86001b6",
      "id": "2e10d8b98127cc06a0f0163e5702",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1704,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.268836498260498,
      "genuineness": 4.770940780639648,
      "wer": 0.0,
      "identity": 0.6356512904167175,
      "voiceclap": -0.022729672491550446,
      "fitness": 0.6039517577087952,
      "legacy_fitness": 0.6039517577087952,
      "target_cdf": 0.49688329211792803,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 1.7544399499893188,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.6209639208598128,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 1.7544399499893188,
          "normalized": 0.49688329211792803,
          "contribution": 0.3478183044825496
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.5174179077148438,
          "normalized": 0.9104853879242107,
          "contribution": 0.2731456163772632
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/2e/2e10d8b98127cc06a0f0163e5702.mp3"
    },
    {
      "metadata_schema": 3,
      "source_metadata_sha256": "c3ab549fcfc681e8e27c66f541e543f1fcb94830c25ca0000c02ac24e3a086fb",
      "id": "9698152fdaafa2594332470c7450",
      "study": "full-v2-sealed",
      "model": "rank4",
      "mission": "whisper_stress",
      "candidate": "sealed-g1",
      "language": "de",
      "reference": true,
      "seed": 1705,
      "valid": true,
      "ended": true,
      "text": "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben. Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte. Mach sie langsam auf und sag mir dann, was du denkst.",
      "sentences": [
        "Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.",
        "Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.",
        "Mach sie langsam auf und sag mir dann, was du denkst."
      ],
      "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Use the quietest intelligible close whisper. Sustain breathy unvoiced delivery with minimal projection, not merely quiet normal speech. Finish every word in the same near-silent whisper. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [6.0 seconds duration] Ich habe diese kleine Schachtel seit dem Tag, an dem wir uns getroffen haben.\n(Natural clear speech.) [5.1 seconds duration] Es ist nichts Teures darin... nur etwas, das ich nicht wegwerfen konnte.\n(Natural clear speech.) [4.7 seconds duration] Mach sie langsam auf und sag mir dann, was du denkst.",
      "requested_seconds": 15.84,
      "actual_seconds": 15.84,
      "cfg": 1.0,
      "generation": null,
      "ce": 6.268836498260498,
      "genuineness": 4.65797233581543,
      "wer": 0.0,
      "identity": 0.3007693290710449,
      "voiceclap": -0.07477593421936035,
      "fitness": 0.47934794800513003,
      "legacy_fitness": 0.47934794800513003,
      "target_cdf": 0.2570897097792827,
      "target_head": "vn_S_WHIS_reg",
      "raw_target": 0.7240524291992188,
      "bursts": [],
      "event_matches": [],
      "family_hit_all": null,
      "nominal_hit_all": null,
      "lexical_proxy": false,
      "cached_historical": false,
      "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
      "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
      "reward_contract": {
        "kind": "emotion-or-style",
        "target_head": "vn_S_WHIS_reg",
        "target_normalization": {
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608
        },
        "weights": {
          "target": 0.5,
          "content_enjoyment": 0.25,
          "genuineness": 0.25
        },
        "normalizers": {
          "content_enjoyment": "clip(CE / 10, 0, 1)",
          "genuineness": "clip(genuineness / 6, 0, 1)",
          "target": "fixed training-distribution Gaussian CDF",
          "blend": "fixed training-distribution Gaussian CDF"
        },
        "wer_factor": "1 - min(1, raw_WER)",
        "family_factor": null,
        "quality_gates": {
          "per_take_wer_max": 0.15,
          "candidate_mean_wer_max": 0.1,
          "valid_terminated_fraction_min": 0.9,
          "reference_identity_delta_min": -0.05,
          "reference_pairs": 50,
          "all_family_fraction_min": null
        },
        "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
      },
      "steering_v2": 0.42918007461988383,
      "steering_components": [
        {
          "head": "vn_S_WHIS_reg",
          "weight": 0.7,
          "direction": "high",
          "mean": 1.7669295072555542,
          "std": 1.5986617803573608,
          "raw": 0.7240524291992188,
          "normalized": 0.2570897097792827,
          "contribution": 0.17996279684549787
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.3,
          "direction": "low",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919,
          "raw": 1.8532664775848389,
          "normalized": 0.8307242592479531,
          "contribution": 0.24921727777438593
        }
      ],
      "steering_contract": {
        "version": "ears-steering-v2",
        "targets": [
          {
            "head": "vn_S_WHIS_reg",
            "weight": 0.7,
            "direction": "high",
            "mean": 1.7669295072555542,
            "std": 1.5986617803573608
          },
          {
            "head": "vn_VOLT_reg",
            "weight": 0.3,
            "direction": "low",
            "mean": 2.6844067573547363,
            "std": 0.8684567809104919
          }
        ],
        "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
        "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
        "content_enjoyment_weight": 0,
        "genuineness_weight": 0,
        "wer_reward_weight": 0,
        "identity_reward_weight": 0,
        "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
        "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
        "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
        "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This 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/96/9698152fdaafa2594332470c7450.mp3"
    }
  ]
}