{
  "schema_version": "1.0",
  "metadata": {
    "evaluation_id": "SYNTHETIC-MODEL-EVAL-001",
    "title": "Synthetic Aggregate Model Evaluation",
    "evaluation_target": "Locked binary research model",
    "target_version": "synthetic-v1",
    "purpose": "Demonstrate aggregate performance and calibration reporting.",
    "population": "Synthetic validation cohort; no real persons or records.",
    "setting": "Synthetic offline evaluation",
    "outcome": "Synthetic binary outcome",
    "outcome_horizon": "Synthetic fixed horizon",
    "threshold": "0.50, prespecified for this synthetic example",
    "threshold_pre_specified": true,
    "threshold_source_id": "SRC-PROTOCOL",
    "validation_dataset": "Independent synthetic aggregate dataset",
    "external_validation": true,
    "data_cut_date": "2026-07-23",
    "data_level": "synthetic",
    "person_level_output": false,
    "human_review_required": true
  },
  "intended_use": "Research evaluation only; not for patient care or live clinical use.",
  "prohibited_uses": [
    "No person-level prediction or classification.",
    "No diagnosis, treatment recommendation, dosing, triage, alarm, or autonomous decision.",
    "No claim of clinical validity, utility, safety, fairness, authorization, or compliance."
  ],
  "limitations": [
    "All counts are synthetic.",
    "Aggregate bins cannot estimate calibration slope.",
    "Subgroup differences are descriptive and do not determine fairness."
  ],
  "human_review": {
    "required": true,
    "roles": [
      "prediction-model methodologist",
      "domain expert",
      "fairness and governance reviewer"
    ],
    "completed": false
  },
  "governance": {
    "owner": "research model governance owner",
    "version_and_change_control": "Lock model, threshold, data, code, and report versions; independently validate every change.",
    "monitoring": "Define aggregate calibration, discrimination, subgroup, drift, failure, and incident review.",
    "auditability": "Retain provenance, commands, versions, reviewer decisions, and unresolved findings without sensitive data.",
    "human_factors": "Document intended users and offline usability evidence; live evaluation is outside this skill.",
    "rollback_and_retirement": "Define rollback, suspension, and retirement triggers before release."
  },
  "sources": [
    {
      "id": "SRC-PROTOCOL",
      "citation": "Synthetic locked evaluation protocol, version 1.",
      "url": "not_applicable_for_local_synthetic_source",
      "accessed": "2026-07-23"
    }
  ],
  "groups": [
    {
      "name": "Synthetic Group A",
      "n": 120,
      "confusion": {
        "tp": 36,
        "fp": 18,
        "tn": 50,
        "fn": 16
      },
      "calibration_bins": [
        {
          "n": 40,
          "mean_predicted_probability": 0.2,
          "observed_events": 12
        },
        {
          "n": 40,
          "mean_predicted_probability": 0.5,
          "observed_events": 17
        },
        {
          "n": 40,
          "mean_predicted_probability": 0.75,
          "observed_events": 23
        }
      ]
    },
    {
      "name": "Synthetic Group B",
      "n": 120,
      "confusion": {
        "tp": 28,
        "fp": 22,
        "tn": 56,
        "fn": 14
      },
      "calibration_bins": [
        {
          "n": 40,
          "mean_predicted_probability": 0.2,
          "observed_events": 11
        },
        {
          "n": 40,
          "mean_predicted_probability": 0.45,
          "observed_events": 14
        },
        {
          "n": 40,
          "mean_predicted_probability": 0.7,
          "observed_events": 17
        }
      ]
    }
  ]
}