{
  "passed": true,
  "rows": 120,
  "calibration_tasks": 30,
  "evaluation_tasks": 90,
  "balanced_evaluation_gold_slots": [
    30,
    30,
    30
  ],
  "calibration_tournament_correct": 29,
  "verified_passing_candidates": 120,
  "verified_failing_candidates": 240,
  "reconstructed_prompt_hashes": 1440,
  "prompt_tokens_min": 105,
  "prompt_tokens_max": 1017,
  "primary_probe_count": 0,
  "diagnostic_probe_count": 270,
  "paired_comparisons": {
    "cyclic": {
      "corrections": 5,
      "regressions": 1,
      "exact_mcnemar_p": 0.21875
    },
    "tournament": {
      "corrections": 3,
      "regressions": 1,
      "exact_mcnemar_p": 0.625
    }
  },
  "note": "Reconciles saved execution evidence and every inference prompt hash; it does not replace the real model run or claim decontamination."
}