{
  "unit": "claim",
  "aggregation": "summed_counts",
  "headline": {
    "metric": "recall",
    "higher_is_better": true,
    "unit": "fraction"
  },
  "secondary": [
    {
      "id": "caught_errors",
      "metric": "count",
      "higher_is_better": true,
      "unit": "claim",
      "aggregation": "median"
    },
    {
      "id": "false_alarms",
      "metric": "count",
      "higher_is_better": false,
      "unit": "claim",
      "aggregation": "median"
    },
    {
      "id": "false_alarm_rate",
      "metric": "recall",
      "higher_is_better": false,
      "unit": "fraction",
      "aggregation": "summed_counts"
    },
    {
      "id": "quote_match_rate",
      "metric": "precision",
      "higher_is_better": true,
      "unit": "fraction",
      "aggregation": "summed_counts"
    },
    {
      "id": "label_accuracy",
      "metric": "accuracy",
      "higher_is_better": true,
      "unit": "fraction",
      "aggregation": "majority_label",
      "labels": [
        "supported",
        "contradicted",
        "not_in_sources",
        "invalid"
      ]
    },
    {
      "id": "invalid_output_rate",
      "metric": "mean",
      "higher_is_better": false,
      "unit": "fraction",
      "aggregation": "median"
    },
    {
      "id": "seeded_errors",
      "metric": "count",
      "higher_is_better": true,
      "unit": "claim",
      "aggregation": "median"
    }
  ],
  "repeats": 3,
  "seed": 421,
  "verdict_rule": {
    "bootstrap": "paired-items",
    "resamples": 10000,
    "confidence": 0.95,
    "prng": "mulberry32",
    "quantile": "type-7",
    "range": "absolute-difference-exceeds-wider-run-range",
    "label_tie": "json-lexicographic",
    "zero_division": 0,
    "degenerate_kappa": 1,
    "rounding": "half-away-from-zero",
    "decimals": 4
  },
  "record_shape": [
    "tp",
    "fp",
    "fn"
  ],
  "prompt": "You verify one draft claim against the supplied sources and nothing else. For each request, one claim in the draft is identified under CHECK ONLY. Decide whether the supplied sources support it (supported), state the opposite (contradicted), or are silent on it (not_in_sources). Quote only exact substrings copied from the supplied sources; when no source supports or contradicts the claim, quote nothing. Do not use outside knowledge and do not judge style. Record the assessment with claim_id set to the requested identifier.",
  "max_tokens": 4096,
  "reasoning_effort": "low",
  "tool_choice": {
    "type": "function",
    "function": {
      "name": "record_claim"
    }
  },
  "planning": {},
  "fixture_sha256": "3cf36a30dd46ea63cfb725b28e8ede51e7010b60d19f254ae92ef6ed67240ab8",
  "recipe_sha256": "12f6306660c3763642b47b5df28b9692de9573379b8b27a11b0ab44074187019",
  "checker_sha256": "f88697af702aad4ccad738fd3653621df649a1c03d94553ec1169978c8eb9c91",
  "runner_sha256": "62c6e3fd116da3dcc6d988e07ddff390b9cefcb51c89b5fbdcd5e25aa41d70e0",
  "fixture_source": {
    "source": "Original synthetic fixture: fictional observatory sources, generated drafts and generator-intended labels",
    "license": "CC0",
    "generators": []
  },
  "sandbox": "off"
}