{
  "schema_version": "clinical_note_audit_paired_generation_v1",
  "study_status": "complete",
  "title": "Fresh paired structured-generation comparison",
  "scope": "Post hoc source-access comparison on previously inspected mock consultations; no clinical efficacy or semantic-truth claim.",
  "design": {
    "consultations": 85,
    "development_consultations": 17,
    "held_consultations": 68,
    "models": [
      "qwen",
      "mistral",
      "phi"
    ],
    "contracts": [
      "scalar",
      "claims"
    ],
    "condition": "source",
    "prompt_only_responses": 510,
    "grammar_responses": 510,
    "scheduled_fresh_responses": 1020,
    "held_pairs": 408,
    "token_cap": 1536,
    "first_completed_attempts_retained": true,
    "fixed_alternating_within_pair_order": true,
    "reference": "Physician risk grade 1 versus grades 2–4; not medical ground truth.",
    "fallback_q": 0.5,
    "bootstrap_samples": 2000,
    "bootstrap_unit": "consultation",
    "bootstrap_seed": 20261023,
    "intervals": "Pointwise exploratory percentile 95%; no multiplicity adjustment or p-values.",
    "historical_archive_is_paired_control": false,
    "historical_replay_exception": "One Phi-4 claims engineering replay differed despite identical inputs; all 510 prompt-only controls were regenerated.",
    "human_results": 0,
    "planned_separate_human_participants": 4
  },
  "macro": {
    "loss": {
      "metric": "loss",
      "family": "macro",
      "style": "macro",
      "consultations": 68,
      "settings_per_consultation": 6,
      "difference_grammar_minus_prompt": 0.015153186274509811,
      "CI95_low": -0.002224264705882344,
      "CI95_high": 0.03294806985294118,
      "pointwise_exploratory": true
    },
    "valid": {
      "metric": "valid",
      "family": "macro",
      "style": "macro",
      "consultations": 68,
      "settings_per_consultation": 6,
      "difference_grammar_minus_prompt": 0.17401960784313725,
      "CI95_low": 0.14215686274509803,
      "CI95_high": 0.20343137254901963,
      "pointwise_exploratory": true
    },
    "schema_EOS_valid": {
      "metric": "schema_EOS_valid",
      "family": "macro",
      "style": "macro",
      "consultations": 68,
      "settings_per_consultation": 6,
      "difference_grammar_minus_prompt": 0.22549019607843135,
      "CI95_low": 0.20343137254901955,
      "CI95_high": 0.24754901960784317,
      "pointwise_exploratory": true
    }
  },
  "groups": [
    {
      "stage": "DEV",
      "family": "mistral",
      "style": "claims",
      "mode": "grammar_constrained",
      "scheduled": 17,
      "valid": 6,
      "coverage": 0.3529411764705882,
      "schema_EOS_valid": 17,
      "pipeline_Brier": 0.2788235294117647,
      "valid_only_Brier": 0.3316666666666667,
      "generated_tokens_mean": 314.5882352941176,
      "new_latency_seconds_median": 11.590378282940948
    },
    {
      "stage": "DEV",
      "family": "mistral",
      "style": "claims",
      "mode": "prompt_only",
      "scheduled": 17,
      "valid": 7,
      "coverage": 0.4117647058823529,
      "schema_EOS_valid": 17,
      "pipeline_Brier": 0.2647058823529412,
      "valid_only_Brier": 0.2857142857142857,
      "generated_tokens_mean": 402.2941176470589,
      "new_latency_seconds_median": 13.965491408016533
    },
    {
      "stage": "DEV",
      "family": "mistral",
      "style": "scalar",
      "mode": "grammar_constrained",
      "scheduled": 17,
      "valid": 17,
      "coverage": 1.0,
      "schema_EOS_valid": 17,
      "pipeline_Brier": 0.3217647058823529,
      "valid_only_Brier": 0.3217647058823529,
      "generated_tokens_mean": 9.0,
      "new_latency_seconds_median": 0.4893482800107449
    },
    {
      "stage": "DEV",
      "family": "mistral",
      "style": "scalar",
      "mode": "prompt_only",
      "scheduled": 17,
      "valid": 17,
      "coverage": 1.0,
      "schema_EOS_valid": 17,
      "pipeline_Brier": 0.3217647058823529,
      "valid_only_Brier": 0.3217647058823529,
      "generated_tokens_mean": 9.0,
      "new_latency_seconds_median": 0.4506186449434608
    },
    {
      "stage": "DEV",
      "family": "phi",
      "style": "claims",
      "mode": "grammar_constrained",
      "scheduled": 17,
      "valid": 15,
      "coverage": 0.8823529411764706,
      "schema_EOS_valid": 17,
      "pipeline_Brier": 0.4945588235294117,
      "valid_only_Brier": 0.5271666666666667,
      "generated_tokens_mean": 306.5882352941176,
      "new_latency_seconds_median": 7.436085355002433
    },
    {
      "stage": "DEV",
      "family": "phi",
      "style": "claims",
      "mode": "prompt_only",
      "scheduled": 17,
      "valid": 1,
      "coverage": 0.0588235294117647,
      "schema_EOS_valid": 1,
      "pipeline_Brier": 0.2829411764705882,
      "valid_only_Brier": 0.81,
      "generated_tokens_mean": 462.1764705882353,
      "new_latency_seconds_median": 15.423654914018698
    },
    {
      "stage": "DEV",
      "family": "phi",
      "style": "scalar",
      "mode": "grammar_constrained",
      "scheduled": 17,
      "valid": 17,
      "coverage": 1.0,
      "schema_EOS_valid": 17,
      "pipeline_Brier": 0.4310294117647059,
      "valid_only_Brier": 0.4310294117647059,
      "generated_tokens_mean": 8.941176470588236,
      "new_latency_seconds_median": 0.3897887879284099
    },
    {
      "stage": "DEV",
      "family": "phi",
      "style": "scalar",
      "mode": "prompt_only",
      "scheduled": 17,
      "valid": 10,
      "coverage": 0.5882352941176471,
      "schema_EOS_valid": 10,
      "pipeline_Brier": 0.3158823529411765,
      "valid_only_Brier": 0.362,
      "generated_tokens_mean": 103.3529411764706,
      "new_latency_seconds_median": 1.137200616998598
    },
    {
      "stage": "DEV",
      "family": "qwen",
      "style": "claims",
      "mode": "grammar_constrained",
      "scheduled": 17,
      "valid": 15,
      "coverage": 0.8823529411764706,
      "schema_EOS_valid": 17,
      "pipeline_Brier": 0.2280882352941176,
      "valid_only_Brier": 0.2251666666666666,
      "generated_tokens_mean": 239.94117647058823,
      "new_latency_seconds_median": 14.39223666896578
    },
    {
      "stage": "DEV",
      "family": "qwen",
      "style": "claims",
      "mode": "prompt_only",
      "scheduled": 17,
      "valid": 14,
      "coverage": 0.8235294117647058,
      "schema_EOS_valid": 17,
      "pipeline_Brier": 0.2354411764705882,
      "valid_only_Brier": 0.2323214285714286,
      "generated_tokens_mean": 315.0,
      "new_latency_seconds_median": 20.15351197496057
    },
    {
      "stage": "DEV",
      "family": "qwen",
      "style": "scalar",
      "mode": "grammar_constrained",
      "scheduled": 17,
      "valid": 17,
      "coverage": 1.0,
      "schema_EOS_valid": 17,
      "pipeline_Brier": 0.190735294117647,
      "valid_only_Brier": 0.190735294117647,
      "generated_tokens_mean": 8.058823529411764,
      "new_latency_seconds_median": 0.7717543379403651
    },
    {
      "stage": "DEV",
      "family": "qwen",
      "style": "scalar",
      "mode": "prompt_only",
      "scheduled": 17,
      "valid": 17,
      "coverage": 1.0,
      "schema_EOS_valid": 17,
      "pipeline_Brier": 0.2483823529411764,
      "valid_only_Brier": 0.2483823529411764,
      "generated_tokens_mean": 12.058823529411764,
      "new_latency_seconds_median": 1.0173111200565472
    },
    {
      "stage": "HOLD",
      "family": "mistral",
      "style": "claims",
      "mode": "grammar_constrained",
      "scheduled": 68,
      "valid": 37,
      "coverage": 0.5441176470588235,
      "schema_EOS_valid": 68,
      "pipeline_Brier": 0.26,
      "valid_only_Brier": 0.2683783783783784,
      "generated_tokens_mean": 259.3529411764706,
      "new_latency_seconds_median": 9.59094757999992
    },
    {
      "stage": "HOLD",
      "family": "mistral",
      "style": "claims",
      "mode": "prompt_only",
      "scheduled": 68,
      "valid": 40,
      "coverage": 0.5882352941176471,
      "schema_EOS_valid": 65,
      "pipeline_Brier": 0.2702941176470588,
      "valid_only_Brier": 0.2845,
      "generated_tokens_mean": 330.72058823529414,
      "new_latency_seconds_median": 11.894432132481596
    },
    {
      "stage": "HOLD",
      "family": "mistral",
      "style": "scalar",
      "mode": "grammar_constrained",
      "scheduled": 68,
      "valid": 68,
      "coverage": 1.0,
      "schema_EOS_valid": 68,
      "pipeline_Brier": 0.2894117647058823,
      "valid_only_Brier": 0.2894117647058823,
      "generated_tokens_mean": 9.0,
      "new_latency_seconds_median": 0.4702869740431197
    },
    {
      "stage": "HOLD",
      "family": "mistral",
      "style": "scalar",
      "mode": "prompt_only",
      "scheduled": 68,
      "valid": 68,
      "coverage": 1.0,
      "schema_EOS_valid": 68,
      "pipeline_Brier": 0.2894117647058823,
      "valid_only_Brier": 0.2894117647058823,
      "generated_tokens_mean": 9.176470588235292,
      "new_latency_seconds_median": 0.4638525729533285
    },
    {
      "stage": "HOLD",
      "family": "phi",
      "style": "claims",
      "mode": "grammar_constrained",
      "scheduled": 68,
      "valid": 50,
      "coverage": 0.7352941176470589,
      "schema_EOS_valid": 68,
      "pipeline_Brier": 0.3047426470588235,
      "valid_only_Brier": 0.32445,
      "generated_tokens_mean": 268.54411764705884,
      "new_latency_seconds_median": 8.993833686516155
    },
    {
      "stage": "HOLD",
      "family": "phi",
      "style": "claims",
      "mode": "prompt_only",
      "scheduled": 68,
      "valid": 0,
      "coverage": 0.0,
      "schema_EOS_valid": 4,
      "pipeline_Brier": 0.25,
      "valid_only_Brier": null,
      "generated_tokens_mean": 398.4411764705882,
      "new_latency_seconds_median": 12.95612103206804
    },
    {
      "stage": "HOLD",
      "family": "phi",
      "style": "scalar",
      "mode": "grammar_constrained",
      "scheduled": 68,
      "valid": 68,
      "coverage": 1.0,
      "schema_EOS_valid": 68,
      "pipeline_Brier": 0.3236764705882353,
      "valid_only_Brier": 0.3236764705882353,
      "generated_tokens_mean": 9.0,
      "new_latency_seconds_median": 0.3866129170055501
    },
    {
      "stage": "HOLD",
      "family": "phi",
      "style": "scalar",
      "mode": "prompt_only",
      "scheduled": 68,
      "valid": 45,
      "coverage": 0.6617647058823529,
      "schema_EOS_valid": 45,
      "pipeline_Brier": 0.2817647058823529,
      "valid_only_Brier": 0.298,
      "generated_tokens_mean": 87.36764705882354,
      "new_latency_seconds_median": 0.6130539989680983
    },
    {
      "stage": "HOLD",
      "family": "qwen",
      "style": "claims",
      "mode": "grammar_constrained",
      "scheduled": 68,
      "valid": 55,
      "coverage": 0.8088235294117647,
      "schema_EOS_valid": 68,
      "pipeline_Brier": 0.2149632352941176,
      "valid_only_Brier": 0.2066818181818181,
      "generated_tokens_mean": 250.16176470588235,
      "new_latency_seconds_median": 17.105731748510152
    },
    {
      "stage": "HOLD",
      "family": "qwen",
      "style": "claims",
      "mode": "prompt_only",
      "scheduled": 68,
      "valid": 54,
      "coverage": 0.7941176470588235,
      "schema_EOS_valid": 66,
      "pipeline_Brier": 0.2276470588235294,
      "valid_only_Brier": 0.2218518518518518,
      "generated_tokens_mean": 352.55882352941177,
      "new_latency_seconds_median": 24.19883803400444
    },
    {
      "stage": "HOLD",
      "family": "qwen",
      "style": "scalar",
      "mode": "grammar_constrained",
      "scheduled": 68,
      "valid": 68,
      "coverage": 1.0,
      "schema_EOS_valid": 68,
      "pipeline_Brier": 0.2605514705882352,
      "valid_only_Brier": 0.2605514705882352,
      "generated_tokens_mean": 8.294117647058824,
      "new_latency_seconds_median": 0.7478511115186848
    },
    {
      "stage": "HOLD",
      "family": "qwen",
      "style": "scalar",
      "mode": "prompt_only",
      "scheduled": 68,
      "valid": 68,
      "coverage": 1.0,
      "schema_EOS_valid": 68,
      "pipeline_Brier": 0.2433088235294117,
      "valid_only_Brier": 0.2433088235294117,
      "generated_tokens_mean": 12.205882352941176,
      "new_latency_seconds_median": 0.9897045365069062
    }
  ],
  "contrasts": [
    {
      "metric": "loss",
      "family": "macro",
      "style": "macro",
      "consultations": 68,
      "settings_per_consultation": 6,
      "difference_grammar_minus_prompt": 0.015153186274509811,
      "CI95_low": -0.002224264705882344,
      "CI95_high": 0.03294806985294118,
      "pointwise_exploratory": true
    },
    {
      "metric": "loss",
      "family": "qwen",
      "style": "scalar",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.017242647058823526,
      "CI95_low": -0.0008097426470588462,
      "CI95_high": 0.040845588235294106,
      "pointwise_exploratory": true
    },
    {
      "metric": "loss",
      "family": "qwen",
      "style": "claims",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": -0.012683823529411753,
      "CI95_low": -0.0604797794117647,
      "CI95_high": 0.032868566176470604,
      "pointwise_exploratory": true
    },
    {
      "metric": "loss",
      "family": "mistral",
      "style": "scalar",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.0,
      "CI95_low": 0.0,
      "CI95_high": 0.0,
      "pointwise_exploratory": true
    },
    {
      "metric": "loss",
      "family": "mistral",
      "style": "claims",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": -0.010294117647058827,
      "CI95_low": -0.0436764705882353,
      "CI95_high": 0.02558823529411764,
      "pointwise_exploratory": true
    },
    {
      "metric": "loss",
      "family": "phi",
      "style": "scalar",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.04191176470588236,
      "CI95_low": -0.010625919117647058,
      "CI95_high": 0.09731709558823529,
      "pointwise_exploratory": true
    },
    {
      "metric": "loss",
      "family": "phi",
      "style": "claims",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.05474264705882354,
      "CI95_low": -0.007611213235294101,
      "CI95_high": 0.12001470588235288,
      "pointwise_exploratory": true
    },
    {
      "metric": "valid",
      "family": "macro",
      "style": "macro",
      "consultations": 68,
      "settings_per_consultation": 6,
      "difference_grammar_minus_prompt": 0.17401960784313725,
      "CI95_low": 0.14215686274509803,
      "CI95_high": 0.20343137254901963,
      "pointwise_exploratory": true
    },
    {
      "metric": "valid",
      "family": "qwen",
      "style": "scalar",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.0,
      "CI95_low": 0.0,
      "CI95_high": 0.0,
      "pointwise_exploratory": true
    },
    {
      "metric": "valid",
      "family": "qwen",
      "style": "claims",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.014705882352941176,
      "CI95_low": -0.04411764705882353,
      "CI95_high": 0.07352941176470588,
      "pointwise_exploratory": true
    },
    {
      "metric": "valid",
      "family": "mistral",
      "style": "scalar",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.0,
      "CI95_low": 0.0,
      "CI95_high": 0.0,
      "pointwise_exploratory": true
    },
    {
      "metric": "valid",
      "family": "mistral",
      "style": "claims",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": -0.04411764705882353,
      "CI95_low": -0.16176470588235295,
      "CI95_high": 0.07352941176470588,
      "pointwise_exploratory": true
    },
    {
      "metric": "valid",
      "family": "phi",
      "style": "scalar",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.3382352941176471,
      "CI95_low": 0.23529411764705882,
      "CI95_high": 0.45588235294117646,
      "pointwise_exploratory": true
    },
    {
      "metric": "valid",
      "family": "phi",
      "style": "claims",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.7352941176470589,
      "CI95_low": 0.6323529411764706,
      "CI95_high": 0.8382352941176471,
      "pointwise_exploratory": true
    },
    {
      "metric": "schema_EOS_valid",
      "family": "macro",
      "style": "macro",
      "consultations": 68,
      "settings_per_consultation": 6,
      "difference_grammar_minus_prompt": 0.22549019607843135,
      "CI95_low": 0.20343137254901955,
      "CI95_high": 0.24754901960784317,
      "pointwise_exploratory": true
    },
    {
      "metric": "schema_EOS_valid",
      "family": "qwen",
      "style": "scalar",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.0,
      "CI95_low": 0.0,
      "CI95_high": 0.0,
      "pointwise_exploratory": true
    },
    {
      "metric": "schema_EOS_valid",
      "family": "qwen",
      "style": "claims",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.029411764705882353,
      "CI95_low": 0.0,
      "CI95_high": 0.07352941176470588,
      "pointwise_exploratory": true
    },
    {
      "metric": "schema_EOS_valid",
      "family": "mistral",
      "style": "scalar",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.0,
      "CI95_low": 0.0,
      "CI95_high": 0.0,
      "pointwise_exploratory": true
    },
    {
      "metric": "schema_EOS_valid",
      "family": "mistral",
      "style": "claims",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.04411764705882353,
      "CI95_low": 0.0,
      "CI95_high": 0.10294117647058823,
      "pointwise_exploratory": true
    },
    {
      "metric": "schema_EOS_valid",
      "family": "phi",
      "style": "scalar",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.3382352941176471,
      "CI95_low": 0.23529411764705882,
      "CI95_high": 0.45588235294117646,
      "pointwise_exploratory": true
    },
    {
      "metric": "schema_EOS_valid",
      "family": "phi",
      "style": "claims",
      "consultations": 68,
      "settings_per_consultation": 1,
      "difference_grammar_minus_prompt": 0.9411764705882353,
      "CI95_low": 0.8823529411764706,
      "CI95_high": 0.9852941176470589,
      "pointwise_exploratory": true
    }
  ],
  "failure_counts": [
    {
      "stage": "DEV",
      "family": "mistral",
      "style": "claims",
      "mode": "grammar_constrained",
      "category": "empty_nonstring_or_nonverbatim_quote",
      "count": 11
    },
    {
      "stage": "DEV",
      "family": "mistral",
      "style": "claims",
      "mode": "grammar_constrained",
      "category": "valid_contract",
      "count": 6
    },
    {
      "stage": "DEV",
      "family": "mistral",
      "style": "claims",
      "mode": "prompt_only",
      "category": "empty_nonstring_or_nonverbatim_quote",
      "count": 10
    },
    {
      "stage": "DEV",
      "family": "mistral",
      "style": "claims",
      "mode": "prompt_only",
      "category": "valid_contract",
      "count": 7
    },
    {
      "stage": "DEV",
      "family": "mistral",
      "style": "scalar",
      "mode": "grammar_constrained",
      "category": "valid_contract",
      "count": 17
    },
    {
      "stage": "DEV",
      "family": "mistral",
      "style": "scalar",
      "mode": "prompt_only",
      "category": "valid_contract",
      "count": 17
    },
    {
      "stage": "DEV",
      "family": "phi",
      "style": "claims",
      "mode": "grammar_constrained",
      "category": "empty_nonstring_or_nonverbatim_quote",
      "count": 2
    },
    {
      "stage": "DEV",
      "family": "phi",
      "style": "claims",
      "mode": "grammar_constrained",
      "category": "valid_contract",
      "count": 15
    },
    {
      "stage": "DEV",
      "family": "phi",
      "style": "claims",
      "mode": "prompt_only",
      "category": "JSON_fence_with_trailing_content",
      "count": 16
    },
    {
      "stage": "DEV",
      "family": "phi",
      "style": "claims",
      "mode": "prompt_only",
      "category": "valid_contract",
      "count": 1
    },
    {
      "stage": "DEV",
      "family": "phi",
      "style": "scalar",
      "mode": "grammar_constrained",
      "category": "valid_contract",
      "count": 17
    },
    {
      "stage": "DEV",
      "family": "phi",
      "style": "scalar",
      "mode": "prompt_only",
      "category": "JSON_fence_with_trailing_content",
      "count": 7
    },
    {
      "stage": "DEV",
      "family": "phi",
      "style": "scalar",
      "mode": "prompt_only",
      "category": "valid_contract",
      "count": 10
    },
    {
      "stage": "DEV",
      "family": "qwen",
      "style": "claims",
      "mode": "grammar_constrained",
      "category": "empty_nonstring_or_nonverbatim_quote",
      "count": 2
    },
    {
      "stage": "DEV",
      "family": "qwen",
      "style": "claims",
      "mode": "grammar_constrained",
      "category": "valid_contract",
      "count": 15
    },
    {
      "stage": "DEV",
      "family": "qwen",
      "style": "claims",
      "mode": "prompt_only",
      "category": "empty_nonstring_or_nonverbatim_quote",
      "count": 3
    },
    {
      "stage": "DEV",
      "family": "qwen",
      "style": "claims",
      "mode": "prompt_only",
      "category": "valid_contract",
      "count": 14
    },
    {
      "stage": "DEV",
      "family": "qwen",
      "style": "scalar",
      "mode": "grammar_constrained",
      "category": "valid_contract",
      "count": 17
    },
    {
      "stage": "DEV",
      "family": "qwen",
      "style": "scalar",
      "mode": "prompt_only",
      "category": "valid_contract",
      "count": 17
    },
    {
      "stage": "HOLD",
      "family": "mistral",
      "style": "claims",
      "mode": "grammar_constrained",
      "category": "empty_nonstring_or_nonverbatim_quote",
      "count": 31
    },
    {
      "stage": "HOLD",
      "family": "mistral",
      "style": "claims",
      "mode": "grammar_constrained",
      "category": "valid_contract",
      "count": 37
    },
    {
      "stage": "HOLD",
      "family": "mistral",
      "style": "claims",
      "mode": "prompt_only",
      "category": "claim_verdict_enum",
      "count": 2
    },
    {
      "stage": "HOLD",
      "family": "mistral",
      "style": "claims",
      "mode": "prompt_only",
      "category": "empty_nonstring_or_nonverbatim_quote",
      "count": 26
    },
    {
      "stage": "HOLD",
      "family": "mistral",
      "style": "claims",
      "mode": "prompt_only",
      "category": "valid_contract",
      "count": 40
    },
    {
      "stage": "HOLD",
      "family": "mistral",
      "style": "scalar",
      "mode": "grammar_constrained",
      "category": "valid_contract",
      "count": 68
    },
    {
      "stage": "HOLD",
      "family": "mistral",
      "style": "scalar",
      "mode": "prompt_only",
      "category": "valid_contract",
      "count": 68
    },
    {
      "stage": "HOLD",
      "family": "phi",
      "style": "claims",
      "mode": "grammar_constrained",
      "category": "empty_nonstring_or_nonverbatim_quote",
      "count": 18
    },
    {
      "stage": "HOLD",
      "family": "phi",
      "style": "claims",
      "mode": "grammar_constrained",
      "category": "valid_contract",
      "count": 50
    },
    {
      "stage": "HOLD",
      "family": "phi",
      "style": "claims",
      "mode": "prompt_only",
      "category": "JSON_fence_with_trailing_content",
      "count": 64
    },
    {
      "stage": "HOLD",
      "family": "phi",
      "style": "claims",
      "mode": "prompt_only",
      "category": "empty_nonstring_or_nonverbatim_quote",
      "count": 4
    },
    {
      "stage": "HOLD",
      "family": "phi",
      "style": "scalar",
      "mode": "grammar_constrained",
      "category": "valid_contract",
      "count": 68
    },
    {
      "stage": "HOLD",
      "family": "phi",
      "style": "scalar",
      "mode": "prompt_only",
      "category": "JSON_fence_with_trailing_content",
      "count": 23
    },
    {
      "stage": "HOLD",
      "family": "phi",
      "style": "scalar",
      "mode": "prompt_only",
      "category": "valid_contract",
      "count": 45
    },
    {
      "stage": "HOLD",
      "family": "qwen",
      "style": "claims",
      "mode": "grammar_constrained",
      "category": "empty_nonstring_or_nonverbatim_quote",
      "count": 13
    },
    {
      "stage": "HOLD",
      "family": "qwen",
      "style": "claims",
      "mode": "grammar_constrained",
      "category": "valid_contract",
      "count": 55
    },
    {
      "stage": "HOLD",
      "family": "qwen",
      "style": "claims",
      "mode": "prompt_only",
      "category": "claim_verdict_enum",
      "count": 1
    },
    {
      "stage": "HOLD",
      "family": "qwen",
      "style": "claims",
      "mode": "prompt_only",
      "category": "empty_nonstring_or_nonverbatim_quote",
      "count": 13
    },
    {
      "stage": "HOLD",
      "family": "qwen",
      "style": "claims",
      "mode": "prompt_only",
      "category": "valid_contract",
      "count": 54
    },
    {
      "stage": "HOLD",
      "family": "qwen",
      "style": "scalar",
      "mode": "grammar_constrained",
      "category": "valid_contract",
      "count": 68
    },
    {
      "stage": "HOLD",
      "family": "qwen",
      "style": "scalar",
      "mode": "prompt_only",
      "category": "valid_contract",
      "count": 68
    }
  ],
  "execution_failures": 0,
  "interpretation": [
    "Coverage measures usability under the frozen quotation-aware contract, not medical correctness.",
    "Pipeline Brier uses every scheduled response, assigning q=0.5 to failures.",
    "Valid-only Brier has a potentially selected denominator and is undefined with zero valid outputs.",
    "The held set was already exposed before this additional intervention; intervals are exploratory.",
    "All six model/contract settings and all DEV/HOLD groups remain visible; no winner selection.",
    "Token and latency summaries describe new runs only, without historical speedup claims.",
    "The four-person user-operated interface evaluation is separate; no human results are included."
  ],
  "provenance": {
    "frozen_protocol_sha256": "2c7d9938ce38df83af5684673553a38e46649b51ea88b5def6d768bccb89a2ce",
    "aggregate_sources_sha256": {
      "PROTOCOL.json": "2c7d9938ce38df83af5684673553a38e46649b51ea88b5def6d768bccb89a2ce",
      "RESULTS.json": "0f70dd627763b66a6960e8537c1512b6da1e130dd7f37b172802d6119f368387",
      "ALL_GROUPS.csv": "ecfe2b8434915b180f75a647d7f99d9351aa57c17b178970857fd30b812c23ee",
      "CONTRASTS.json": "3ff8121241b98e47f85c0af4791ef27594a954a00f1c59bde864e35505064bb2",
      "ALL_FORECASTS.csv": "da7e3e7b89647a8b9eaf41963614a3deb1024aa018bf457b8db5d9424256a909",
      "ALL_FAILURES.csv": "fcf418f880db01c5a3fe0805aa89e8aa98a248341daa162ff8eda5a2079fb1db",
      "ANALYSIS_VALIDATION.json": "534e19d283966ea619871748a517aecd207e9480594acc23a068fb30ee2507d8"
    },
    "independent_recomputation": "pass",
    "reference_labels_loaded_by_generator": false,
    "new_training": false,
    "individual_responses_in_this_export": false,
    "raw_precision_retained": true,
    "protocol_version": "B_structured_decoding_fresh_paired_v2"
  }
}
