{
  "report_id": "f7-evaluator-metaeval-reference",
  "chapter": "07.04",
  "rubric_version": "support-answer-rubric-v3",
  "reference_cases": 80,
  "metrics": {
    "agreement": 0.82,
    "weighted_kappa": 0.71,
    "undue_pass_rate": 0.06,
    "undue_fail_rate": 0.09,
    "borderline_overturn_rate": 0.16,
    "rubric_coverage": 0.93
  },
  "calibration_notes": [
    "Los casos frontera se revisan con doble lectura humana antes de ajustar la rúbrica.",
    "El evaluador no decide release: aporta una señal versionada dentro del expediente."
  ]
}
