{
  "status": "Executed exact synthetic diagnostics; no LLM benchmark",
  "scipy_version": "1.17.0",
  "alpha": 0.02,
  "delta": 0.05,
  "zero_error_bounds": [
    {
      "accepted": 50,
      "candidates": 1,
      "upper": 0.05815507911697225
    },
    {
      "accepted": 149,
      "candidates": 1,
      "upper": 0.019904816220685077
    },
    {
      "accepted": 300,
      "candidates": 1,
      "upper": 0.009936081944457708
    },
    {
      "accepted": 300,
      "candidates": 20,
      "upper": 0.019773438164508515
    }
  ],
  "minimum_zero_error_acceptances": {
    "1": 149,
    "3": 203,
    "20": 297,
    "160": 400
  },
  "worked_calibration": [
    {
      "threshold": 0.8,
      "accepted": 1000,
      "errors": 30,
      "coverage": 1.0,
      "empirical_risk": 0.03,
      "p_value": 0.9873522516538874,
      "upper_bound": 0.04368693745549481,
      "certified": false
    },
    {
      "threshold": 0.95,
      "accepted": 600,
      "errors": 4,
      "coverage": 0.6,
      "empirical_risk": 0.006666666666666667,
      "p_value": 0.007181221123099641,
      "upper_bound": 0.01798641315508748,
      "certified": true
    },
    {
      "threshold": 0.99,
      "accepted": 300,
      "errors": 0,
      "coverage": 0.3,
      "empirical_risk": 0.0,
      "p_value": 0.0023325056679514245,
      "upper_bound": 0.013555106015980924,
      "certified": true
    }
  ],
  "selected_threshold": 0.95,
  "multiplicity_counterexample": [
    {
      "method": "uncorrected_search",
      "n": 300,
      "candidates": 20,
      "true_risk": 0.0201,
      "max_errors": 1,
      "single_false_certificate_probability": 0.016182937813510923,
      "family_false_certificate_probability": 0.27841581894611556
    },
    {
      "method": "bonferroni",
      "n": 300,
      "candidates": 20,
      "true_risk": 0.0201,
      "max_errors": 0,
      "single_false_certificate_probability": 0.002262180739163095,
      "family_false_certificate_probability": 0.044284368441907196
    }
  ],
  "risk_denominators": {
    "tasks": 1000,
    "released": 100,
    "wrong_released": 20,
    "unconditional_wrong_release_rate": 0.02,
    "selective_risk": 0.2,
    "coverage": 0.1,
    "mean_FDP_if_every_answer_has_1_error_in_100_claims": 0.01,
    "any_error_answer_rate_in_same_fixture": 1.0
  },
  "threshold_nonmonotonicity": {
    "two_claim_FDP_before": 0.5,
    "after_rejecting_correct_claim": 1.0,
    "answer_risk_at_score_threshold_0_8": 0.02,
    "answer_risk_at_score_threshold_0_99": 0.2
  },
  "dependent_duplicates": {
    "independent_families": 10,
    "copies_per_family": 100,
    "true_risk": 0.1,
    "actual_zero_error_probability": 0.3486784401000001,
    "incorrect_iid_zero_error_probability": 1.7478712517226947e-46,
    "naive_upper_from_1000_zero_errors": 0.002991249545095295,
    "family_upper_from_10_zero_errors": 0.2588655508930522
  },
  "pooled_vs_revised": {
    "unchanged": {
      "n": 900,
      "errors": 9
    },
    "revised": {
      "n": 100,
      "errors": 20
    },
    "pooled_error": 0.029,
    "revised_error": 0.2
  },
  "structural_guard_checks": [
    {
      "fixture": "valid",
      "released": true,
      "reason": "structural checks passed",
      "passed": true
    },
    {
      "fixture": "changed_answer",
      "released": false,
      "reason": "answer changed after scoring",
      "passed": true
    },
    {
      "fixture": "changed_pipeline",
      "released": false,
      "reason": "pipeline_digest mismatch",
      "passed": true
    },
    {
      "fixture": "changed_score",
      "released": false,
      "reason": "score_digest mismatch",
      "passed": true
    },
    {
      "fixture": "changed_sources",
      "released": false,
      "reason": "evidence_policy_digest mismatch",
      "passed": true
    },
    {
      "fixture": "changed_population",
      "released": false,
      "reason": "population_id mismatch",
      "passed": true
    },
    {
      "fixture": "changed_label_contract",
      "released": false,
      "reason": "label_contract_digest mismatch",
      "passed": true
    },
    {
      "fixture": "missing_usage",
      "released": false,
      "reason": "incomplete usage",
      "passed": true
    },
    {
      "fixture": "incomplete_answer",
      "released": false,
      "reason": "incomplete answer or evidence",
      "passed": true
    },
    {
      "fixture": "below_threshold",
      "released": false,
      "reason": "below threshold",
      "passed": true
    }
  ],
  "fixture_notes": [
    "All scores, labels, costs, and answers are invented synthetic fixtures.",
    "Digest matching cannot detect an unrecorded provider update or false provenance.",
    "Counting families is not proof that distinct families are independent.",
    "An unavailable reasoning-token count must be recorded as unknown in real studies; this guard requires a complete accounting adapter and otherwise cannot claim exact compute matching."
  ]
}