{
  "schema_version": 1,
  "columns": [
    "configuration",
    "threshold",
    "metric",
    "retained_answers",
    "estimate",
    "ci_low",
    "ci_high",
    "valid_resamples",
    "interval_status"
  ],
  "rows": [
    {
      "configuration": "Flash-Lite MINIMAL corrected",
      "threshold": "high",
      "metric": "coverage",
      "retained_answers": "387",
      "estimate": "0.489873417721519",
      "ci_low": "0.4538651002077612",
      "ci_high": "0.5262484395006402",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite MINIMAL corrected",
      "threshold": "high",
      "metric": "error_among_retained",
      "retained_answers": "387",
      "estimate": "0.19121447028423777",
      "ci_low": "0.15308578987150417",
      "ci_high": "0.23184357541899442",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite MINIMAL corrected",
      "threshold": "medium_or_high",
      "metric": "coverage",
      "retained_answers": "481",
      "estimate": "0.6088607594936709",
      "ci_low": "0.5730619016832086",
      "ci_high": "0.6436640995687852",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite MINIMAL corrected",
      "threshold": "medium_or_high",
      "metric": "error_among_retained",
      "retained_answers": "481",
      "estimate": "0.2577962577962578",
      "ci_low": "0.2183655437226866",
      "ci_high": "0.2981366459627329",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite MINIMAL corrected",
      "threshold": "all_reported",
      "metric": "coverage",
      "retained_answers": "482",
      "estimate": "0.610126582278481",
      "ci_low": "0.5746835443037974",
      "ci_high": "0.6448729611041405",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite MINIMAL corrected",
      "threshold": "all_reported",
      "metric": "error_among_retained",
      "retained_answers": "482",
      "estimate": "0.2572614107883817",
      "ci_low": "0.21792059984281378",
      "ci_high": "0.2975726924377324",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite LOW",
      "threshold": "high",
      "metric": "coverage",
      "retained_answers": "365",
      "estimate": "0.4620253164556962",
      "ci_low": "0.425531914893617",
      "ci_high": "0.4987244897959184",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite LOW",
      "threshold": "high",
      "metric": "error_among_retained",
      "retained_answers": "365",
      "estimate": "0.16164383561643836",
      "ci_low": "0.12532637075718014",
      "ci_high": "0.2",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite LOW",
      "threshold": "medium_or_high",
      "metric": "coverage",
      "retained_answers": "488",
      "estimate": "0.6177215189873417",
      "ci_low": "0.5818402918683918",
      "ci_high": "0.6525002413127413",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite LOW",
      "threshold": "medium_or_high",
      "metric": "error_among_retained",
      "retained_answers": "488",
      "estimate": "0.24795081967213117",
      "ci_low": "0.20939922480620154",
      "ci_high": "0.288271279580581",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite LOW",
      "threshold": "all_reported",
      "metric": "coverage",
      "retained_answers": "496",
      "estimate": "0.6278481012658228",
      "ci_low": "0.5919395465994962",
      "ci_high": "0.6629076201740306",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite LOW",
      "threshold": "all_reported",
      "metric": "error_among_retained",
      "retained_answers": "496",
      "estimate": "0.25806451612903225",
      "ci_low": "0.21923016701461379",
      "ci_high": "0.2986254168701334",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite MEDIUM",
      "threshold": "high",
      "metric": "coverage",
      "retained_answers": "346",
      "estimate": "0.4379746835443038",
      "ci_low": "0.4022364447570973",
      "ci_high": "0.4742268041237113",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite MEDIUM",
      "threshold": "high",
      "metric": "error_among_retained",
      "retained_answers": "346",
      "estimate": "0.14161849710982655",
      "ci_low": "0.10542168674698796",
      "ci_high": "0.1802337864395676",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite MEDIUM",
      "threshold": "medium_or_high",
      "metric": "coverage",
      "retained_answers": "545",
      "estimate": "0.689873417721519",
      "ci_low": "0.6563246246003157",
      "ci_high": "0.7229987293519695",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite MEDIUM",
      "threshold": "medium_or_high",
      "metric": "error_among_retained",
      "retained_answers": "545",
      "estimate": "0.26788990825688075",
      "ci_low": "0.2300556586270872",
      "ci_high": "0.30666666666666664",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite MEDIUM",
      "threshold": "all_reported",
      "metric": "coverage",
      "retained_answers": "545",
      "estimate": "0.689873417721519",
      "ci_low": "0.6563246246003157",
      "ci_high": "0.7229987293519695",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite MEDIUM",
      "threshold": "all_reported",
      "metric": "error_among_retained",
      "retained_answers": "545",
      "estimate": "0.26788990825688075",
      "ci_low": "0.2300556586270872",
      "ci_high": "0.30666666666666664",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite HIGH",
      "threshold": "high",
      "metric": "coverage",
      "retained_answers": "387",
      "estimate": "0.489873417721519",
      "ci_low": "0.45387430869321166",
      "ci_high": "0.5259069813903827",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite HIGH",
      "threshold": "high",
      "metric": "error_among_retained",
      "retained_answers": "387",
      "estimate": "0.17829457364341084",
      "ci_low": "0.14077580007191656",
      "ci_high": "0.21787888554648327",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite HIGH",
      "threshold": "medium_or_high",
      "metric": "coverage",
      "retained_answers": "578",
      "estimate": "0.7316455696202532",
      "ci_low": "0.6997455470737913",
      "ci_high": "0.7624033392355172",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite HIGH",
      "threshold": "medium_or_high",
      "metric": "error_among_retained",
      "retained_answers": "578",
      "estimate": "0.2975778546712803",
      "ci_low": "0.25925925925925924",
      "ci_high": "0.33668469900523024",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite HIGH",
      "threshold": "all_reported",
      "metric": "coverage",
      "retained_answers": "578",
      "estimate": "0.7316455696202532",
      "ci_low": "0.6997455470737913",
      "ci_high": "0.7624033392355172",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Flash-Lite HIGH",
      "threshold": "all_reported",
      "metric": "error_among_retained",
      "retained_answers": "578",
      "estimate": "0.2975778546712803",
      "ci_low": "0.25925925925925924",
      "ci_high": "0.33668469900523024",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.7 MEDIUM media / MEDIUM thinking",
      "threshold": "high",
      "metric": "coverage",
      "retained_answers": "489",
      "estimate": "0.6189873417721519",
      "ci_low": "0.5833333333333334",
      "ci_high": "0.6533174925486755",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.7 MEDIUM media / MEDIUM thinking",
      "threshold": "high",
      "metric": "error_among_retained",
      "retained_answers": "489",
      "estimate": "0.1492842535787321",
      "ci_low": "0.11847389558232932",
      "ci_high": "0.18217841824668296",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.7 MEDIUM media / MEDIUM thinking",
      "threshold": "medium_or_high",
      "metric": "coverage",
      "retained_answers": "527",
      "estimate": "0.6670886075949367",
      "ci_low": "0.6322971348427169",
      "ci_high": "0.7009928675636246",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.7 MEDIUM media / MEDIUM thinking",
      "threshold": "medium_or_high",
      "metric": "error_among_retained",
      "retained_answers": "527",
      "estimate": "0.16508538899430736",
      "ci_low": "0.1342155009451796",
      "ci_high": "0.1981485728848114",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.7 MEDIUM media / MEDIUM thinking",
      "threshold": "all_reported",
      "metric": "coverage",
      "retained_answers": "529",
      "estimate": "0.669620253164557",
      "ci_low": "0.6348802842082747",
      "ci_high": "0.703466181112771",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.7 MEDIUM media / MEDIUM thinking",
      "threshold": "all_reported",
      "metric": "error_among_retained",
      "retained_answers": "529",
      "estimate": "0.16824196597353502",
      "ci_low": "0.13725490196078433",
      "ci_high": "0.20156555772994128",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / MEDIUM thinking",
      "threshold": "high",
      "metric": "coverage",
      "retained_answers": "461",
      "estimate": "0.5835443037974684",
      "ci_low": "0.5480991183001346",
      "ci_high": "0.6186240594722737",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / MEDIUM thinking",
      "threshold": "high",
      "metric": "error_among_retained",
      "retained_answers": "461",
      "estimate": "0.11930585683297179",
      "ci_low": "0.0904936974789916",
      "ci_high": "0.14989293361884368",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / MEDIUM thinking",
      "threshold": "medium_or_high",
      "metric": "coverage",
      "retained_answers": "522",
      "estimate": "0.660759493670886",
      "ci_low": "0.6262886597938144",
      "ci_high": "0.695201364317988",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / MEDIUM thinking",
      "threshold": "medium_or_high",
      "metric": "error_among_retained",
      "retained_answers": "522",
      "estimate": "0.14176245210727967",
      "ci_low": "0.11222347984169499",
      "ci_high": "0.1733074369189907",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / MEDIUM thinking",
      "threshold": "all_reported",
      "metric": "coverage",
      "retained_answers": "522",
      "estimate": "0.660759493670886",
      "ci_low": "0.6262886597938144",
      "ci_high": "0.695201364317988",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / MEDIUM thinking",
      "threshold": "all_reported",
      "metric": "error_among_retained",
      "retained_answers": "522",
      "estimate": "0.14176245210727967",
      "ci_low": "0.11222347984169499",
      "ci_high": "0.1733074369189907",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / LOW thinking",
      "threshold": "high",
      "metric": "coverage",
      "retained_answers": "470",
      "estimate": "0.5949367088607594",
      "ci_low": "0.558934408264006",
      "ci_high": "0.6305732484076433",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / LOW thinking",
      "threshold": "high",
      "metric": "error_among_retained",
      "retained_answers": "470",
      "estimate": "0.11914893617021272",
      "ci_low": "0.09071729957805907",
      "ci_high": "0.14935064935064934",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / LOW thinking",
      "threshold": "medium_or_high",
      "metric": "coverage",
      "retained_answers": "524",
      "estimate": "0.6632911392405063",
      "ci_low": "0.628463476070529",
      "ci_high": "0.6977048045288644",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / LOW thinking",
      "threshold": "medium_or_high",
      "metric": "error_among_retained",
      "retained_answers": "524",
      "estimate": "0.14122137404580148",
      "ci_low": "0.11216685618847945",
      "ci_high": "0.17190838790851862",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / LOW thinking",
      "threshold": "all_reported",
      "metric": "coverage",
      "retained_answers": "527",
      "estimate": "0.6670886075949367",
      "ci_low": "0.6326276463262764",
      "ci_high": "0.7013801756587202",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / LOW thinking",
      "threshold": "all_reported",
      "metric": "error_among_retained",
      "retained_answers": "527",
      "estimate": "0.1461100569259962",
      "ci_low": "0.11706195046574389",
      "ci_high": "0.17729183813193294",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / HIGH thinking",
      "threshold": "high",
      "metric": "coverage",
      "retained_answers": "495",
      "estimate": "0.6265822784810127",
      "ci_low": "0.591304347826087",
      "ci_high": "0.661558597291685",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / HIGH thinking",
      "threshold": "high",
      "metric": "error_among_retained",
      "retained_answers": "495",
      "estimate": "0.12525252525252528",
      "ci_low": "0.0973630831643002",
      "ci_high": "0.1558718131086552",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / HIGH thinking",
      "threshold": "medium_or_high",
      "metric": "coverage",
      "retained_answers": "611",
      "estimate": "0.7734177215189874",
      "ci_low": "0.7433619885079444",
      "ci_high": "0.8032388787946838",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / HIGH thinking",
      "threshold": "medium_or_high",
      "metric": "error_among_retained",
      "retained_answers": "611",
      "estimate": "0.1718494271685761",
      "ci_low": "0.14214876033057852",
      "ci_high": "0.20333403361344538",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / HIGH thinking",
      "threshold": "all_reported",
      "metric": "coverage",
      "retained_answers": "611",
      "estimate": "0.7734177215189874",
      "ci_low": "0.7433619885079444",
      "ci_high": "0.8032388787946838",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Gemini 3.8 HIGH media / HIGH thinking",
      "threshold": "all_reported",
      "metric": "error_among_retained",
      "retained_answers": "611",
      "estimate": "0.1718494271685761",
      "ci_low": "0.14214876033057852",
      "ci_high": "0.20333403361344538",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Qwen Instruct first_pass",
      "threshold": "high",
      "metric": "coverage",
      "retained_answers": "447",
      "estimate": "0.5658227848101266",
      "ci_low": "0.5299984076433121",
      "ci_high": "0.6017811704834606",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Qwen Instruct first_pass",
      "threshold": "high",
      "metric": "error_among_retained",
      "retained_answers": "447",
      "estimate": "0.35794183445190153",
      "ci_low": "0.3136781732418525",
      "ci_high": "0.4048140043763676",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Qwen Instruct first_pass",
      "threshold": "medium_or_high",
      "metric": "coverage",
      "retained_answers": "667",
      "estimate": "0.8443037974683544",
      "ci_low": "0.8176020408163265",
      "ci_high": "0.8697318007662835",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Qwen Instruct first_pass",
      "threshold": "medium_or_high",
      "metric": "error_among_retained",
      "retained_answers": "667",
      "estimate": "0.5217391304347826",
      "ci_low": "0.4817073170731707",
      "ci_high": "0.561482905982906",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Qwen Instruct first_pass",
      "threshold": "all_reported",
      "metric": "coverage",
      "retained_answers": "676",
      "estimate": "0.8556962025316456",
      "ci_low": "0.8292993630573249",
      "ci_high": "0.8804881883724498",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Qwen Instruct first_pass",
      "threshold": "all_reported",
      "metric": "error_among_retained",
      "retained_answers": "676",
      "estimate": "0.5281065088757396",
      "ci_low": "0.48805880775761906",
      "ci_high": "0.56732223903177",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Qwen Instruct primary_repaired",
      "threshold": "high",
      "metric": "coverage",
      "retained_answers": "447",
      "estimate": "0.5658227848101266",
      "ci_low": "0.5299984076433121",
      "ci_high": "0.6017811704834606",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Qwen Instruct primary_repaired",
      "threshold": "high",
      "metric": "error_among_retained",
      "retained_answers": "447",
      "estimate": "0.35794183445190153",
      "ci_low": "0.3136781732418525",
      "ci_high": "0.4048140043763676",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Qwen Instruct primary_repaired",
      "threshold": "medium_or_high",
      "metric": "coverage",
      "retained_answers": "669",
      "estimate": "0.8468354430379746",
      "ci_low": "0.820351019347045",
      "ci_high": "0.8716989355952912",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Qwen Instruct primary_repaired",
      "threshold": "medium_or_high",
      "metric": "error_among_retained",
      "retained_answers": "669",
      "estimate": "0.5216741405082213",
      "ci_low": "0.48192634586685107",
      "ci_high": "0.5614814814814815",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Qwen Instruct primary_repaired",
      "threshold": "all_reported",
      "metric": "coverage",
      "retained_answers": "680",
      "estimate": "0.8607594936708861",
      "ci_low": "0.8350122795969774",
      "ci_high": "0.88510101010101",
      "valid_resamples": "10000",
      "interval_status": "ok"
    },
    {
      "configuration": "Qwen Instruct primary_repaired",
      "threshold": "all_reported",
      "metric": "error_among_retained",
      "retained_answers": "680",
      "estimate": "0.5294117647058824",
      "ci_low": "0.4899706262556401",
      "ci_high": "0.5689149560117303",
      "valid_resamples": "10000",
      "interval_status": "ok"
    }
  ],
  "unit_policy": {
    "table": "selective_confidence.csv",
    "metric_units": {
      "coverage": "proportion_0_to_1",
      "error_among_retained": "proportion_0_to_1"
    },
    "counts": {
      "retained_answers": "answers",
      "valid_resamples": "bootstrap_resamples"
    },
    "interval": "ci_low and ci_high use the same units as estimate; pointwise 95% when interval_status is ok",
    "estimate": "unit selected by coverage/error_among_retained metric_units entry",
    "scope": "verbal-confidence threshold selection; not calibrated probability"
  },
  "note": "Saved aggregate candidate analysis; intervals are pointwise and exploratory tests have separate Holm-adjusted p-values."
}