{
 "accuracy_vs_ground_truth": 0.3937,
 "basis": "3600 s actual elapsed in one process; 80 cycles; 640 governed tasks (560 local model calls, 80 sandbox executions); 0 failures; 0 duplicate task ids; checkpoint chain intact; duty-cycled (44% busy); accuracy 252/640 is a small-model score, not a quality claim",
 "checkpoint_chain_ok": true,
 "checkpoints": 80,
 "clock": "ACTUAL elapsed wall clock; no virtual or accelerated time",
 "cloud_calls": 0,
 "cost_usd": 0.0,
 "cpu_seconds_user_plus_system": 2991.4,
 "duplicate_task_ids": 0,
 "duty_cycle": {
  "busy_fraction": 0.4359,
  "busy_s": 1569.4,
  "note": "elapsed time is not productive time; the campaign sleeps between cycles",
  "period_s": 45.0
 },
 "elapsed_s": 3600.0,
 "end_utc": 1794787125.998,
 "execution_classes": [
  "LOCAL_MODEL_INFERENCE",
  "DETERMINISTIC_EXECUTION"
 ],
 "failures": 0,
 "gate_audit_chain_ok": true,
 "gate_events": 1306,
 "host": "one host, one process, shared with production services",
 "kill_switch": {
  "decision_after_kill_refused": true,
  "engage_ms_in_process": 0.407,
  "scope": "in-process gate only; no distributed halt"
 },
 "loadavg_1m": {
  "max": 1705.77,
  "mean": 56.02,
  "min": 7.09
 },
 "model": "qwen2.5-0.5b-instruct",
 "model_calls": 560,
 "model_calls_with_grant_id": 560,
 "not_evaluated": [
  "6 h, 24 h, 72 h, 168 h campaigns",
  "failure injection during the campaign",
  "cross-host execution",
  "storage and network metrics"
 ],
 "peak_rss_mb": 1625.1,
 "requested_s": 3600.0,
 "retries": 0,
 "schema": "cain2027.e80.endurance.v1",
 "scoring": "ground truth computed by task-generator code; not independent agent verification",
 "start_utc": 1794783525.998,
 "status": "PASS",
 "threads": 2,
 "tokens": 63794,
 "totals": {
  "correct": 252,
  "cycles": 80,
  "errors": 0,
  "model_tasks": 560,
  "overlap_with_benchmark_questions": 109,
  "replay_attempts": 8,
  "replay_refused": 8,
  "sandbox_tasks": 80,
  "tasks": 640,
  "unparsed": 0
 }
}