{
  "schema_version": "agent-benchmark-spec-v1",
  "benchmark_id": "readonly-order-agent",
  "benchmark_version": "1.0.0",
  "private_test_frozen": true,
  "claim": "Compare complete readonly order-agent systems under one frozen protocol.",
  "target_population": "Synthetic Chinese readonly order-support tasks for teaching.",
  "baseline_system_id": "baseline-v1",
  "primary_metric": "task_success_rate",
  "protocol": {
    "protocol_version": "protocol-v1",
    "environment_id": "offline-order-fixture-v1",
    "toolset_id": "readonly-tools-v1",
    "reset_policy": "reset-before-every-trial",
    "max_steps": 9,
    "timeout_seconds": 30,
    "retry_limit": 0,
    "trial_count": 3
  },
  "gates": {
    "task_min_trial_success_rate": 1.0,
    "min_primary_task_success_rate": 0.8,
    "critical_task_min_trial_success_rate": 1.0,
    "min_final_state_trial_rate": 0.8,
    "min_side_effect_free_trial_rate": 0.8,
    "min_trial_stability_rate": 0.8,
    "max_slice_task_success_gap": 0.5,
    "max_mean_latency_ms": 1200,
    "max_p95_latency_ms": 1500,
    "max_mean_cost_units": 2.0,
    "min_paired_delta_ci_low": -0.2
  },
  "bootstrap": {
    "samples": 2000,
    "seed": 23,
    "confidence": 0.95
  }
}