{
  "audit_id": "d2_diagonal_20261027_independent_aggregation",
  "actor": "/root/protocol_audit",
  "actor_type": "agent",
  "audited_at": "2026-10-27T12:05:20.719586+08:00",
  "status": "passed",
  "scope": "Read-only verification of saved outcomes; no new experiment, training or replacement of primary results.",
  "source_contract_commit": "d2796a484712fc95268c2400a6d85d5cf121163c",
  "audited_artifact_sha256": {
    "rows.jsonl": "362452a7872c40c40ef55168fc7db7eb2d32a24f9c76d5deebd6cc55513467c4",
    "summary.json": "6267405df4a675eff65297f94e3c9563d5c84b18f3c543628367ae2bdac737eb",
    "execution.json": "41d82b6f878b069491c31a756f7118d434cd6850b63a1a518c02b57bfdfa035e",
    "input_verification.json": "d8aeed004621ece707687cbef86b003ae66b08cf381f5c19c2b0b2fb1a4b5a4f",
    "numerical_verification.json": "87fb9f8c1eef853b4eaacd963c9da3fb098d3c69bf5b6ed50c1971e8220d7130",
    "preregistration.json": "1d93ba4bc34d733f8ca04304112be82631fa937a8d48489d91a2f8e04f370eb4"
  },
  "frozen_source_hashes_match": true,
  "row_checks": {
    "status": "passed",
    "rows": 3240,
    "unique_curves": 3240,
    "unique_draws": 120,
    "interior": 3137,
    "boundary": 103,
    "predicates": [
      "time_factor",
      "time_pass",
      "regret_raw",
      "regret_over_initial",
      "risk_pass",
      "invalid_bias",
      "invalid_total"
    ]
  },
  "aggregation_checks": {
    "status": "passed",
    "group_count": 52,
    "composition": {
      "overall": 1,
      "recipe": 12,
      "seed": 3,
      "recipe_seed": 36
    },
    "verified_fields": [
      "count",
      "interior",
      "boundary",
      "time_pass",
      "time_rate",
      "risk_pass",
      "risk_rate",
      "stopping_pass",
      "invalid_bias",
      "invalid_total",
      "valid_and_risk_pass",
      "predicted_boundary",
      "boundary_status_agreement",
      "baseline_time_pass",
      "baseline_risk_pass",
      "baseline_clipped"
    ],
    "overall_quantiles_verified": [
      "regret_over_initial",
      "time_factor_positive_predictions"
    ]
  },
  "primary_conclusion": {
    "required_recipes_passing": 12,
    "observed_recipes_passing": 6,
    "status": "refuted_on_development",
    "time_pass": 2352,
    "time_denominator": 3137,
    "risk_pass": 2739,
    "risk_denominator": 3240
  },
  "validity_not_equivalent_to_stopping_quality": {
    "invalid_bias": 693,
    "invalid_bias_and_risk_pass": 642,
    "invalid_total": 279,
    "invalid_total_and_risk_pass": 253,
    "worst_regret_has_valid_bias_and_total": true,
    "worst_regret_example": {
      "protocol": "d1_biased",
      "task": "sine",
      "n": 32,
      "seed": 6103,
      "sigma2": 1,
      "t_reference": 255,
      "t_diagonal": 1,
      "risk_reference": 0.25205198134278284,
      "full_risk_at_diagonal": 0.8546996829218976,
      "regret_over_initial": 0.5884719854443935
    }
  },
  "seed_results": [
    {
      "seed": 6101,
      "time_pass": 854,
      "interior": 1044,
      "time_rate": 0.8180076628352491,
      "risk_pass": 1007,
      "count": 1080,
      "risk_rate": 0.9324074074074075,
      "recipes_passing_both": 8
    },
    {
      "seed": 6102,
      "time_pass": 742,
      "interior": 1045,
      "time_rate": 0.7100478468899522,
      "risk_pass": 878,
      "count": 1080,
      "risk_rate": 0.812962962962963,
      "recipes_passing_both": 2
    },
    {
      "seed": 6103,
      "time_pass": 756,
      "interior": 1048,
      "time_rate": 0.7213740458015268,
      "risk_pass": 854,
      "count": 1080,
      "risk_rate": 0.7907407407407407,
      "recipes_passing_both": 4
    }
  ],
  "recipes_passing_all_three_seeds": [
    {
      "protocol": "d4_homogeneous",
      "task": "sine"
    }
  ],
  "negative_raw_regret": {
    "count": 295,
    "minimum": -1.1823007850519929e-14,
    "maximum": -8.673617379884035e-19,
    "all_within_registered_roundoff_tolerance": true,
    "genuine_search_discrepancies": 0
  },
  "tail_summary": {
    "regret_over_initial": {
      "min": 0.0,
      "p25": 5.022656878452566e-06,
      "median": 0.0006741145240927659,
      "p75": 0.00433111782786417,
      "p90": 0.017794652241004592,
      "p95": 0.03421647762662494,
      "max": 0.5884719854443935
    },
    "time_factor_positive_predictions": {
      "min": 1.0,
      "p25": 1.050840025846949,
      "median": 1.2595108695652173,
      "p75": 2.0,
      "p90": 8.039615846338535,
      "p95": 26.15219473264166,
      "max": 1584.6785714285713
    },
    "zero_time_predictions_excluded_from_factor_quantiles": 6
  },
  "interpretation_limits": [
    "Development data only.",
    "693 negative signal-risk curves do not imply 693 bad stopping choices.",
    "6/12 pooled recipe successes do not mean six recipes passed for each seed.",
    "No binomial confidence interval over correlated curves; only three independent master seeds.",
    "Near-degenerate eigenbasis convention remains a registered sensitivity limitation."
  ]
}
