docs/results/improvement_report_benchmark.json

{
  "status": "ok",
  "baseline": "outputs/reports/baselines.json",
  "candidate": "outputs/reports/benchmark_report.json",
  "deltas": {
    "avg_reward": -0.0025,
    "legality_rate": 0.0,
    "success_rate": 0.0,
    "avg_process_fidelity": 0.92,
    "timeout_rate": 0.0,
    "failure_visible_rate": 0.0
  },
  "gate": {
    "avg_reward_up": false,
    "legality_up": true,
    "success_up": true
  },
  "improved": false
}