{
  "model": "Secret-700-SFT",
  "harness": "deepagents",
  "status": "completed",
  "captured_at": "2026-10-18T07:50:54Z",
  "task_set_version": "0.18.0",
  "total_tasks": 431,
  "completed_tasks": 431,
  "passed_tasks": 377,
  "failed_tasks": 54,
  "pending_tasks": 0,
  "infrastructure_failures": 0,
  "attempts_per_task": 1,
  "reasoning": "enabled; effort default; reasoning history forwarded",
  "max_output_tokens": 81920,
  "temperature": 0.8,
  "top_p": 0.95,
  "concurrency": 2,
  "isolation": "Linux bwrap",
  "harness_version": "0.6.12",
  "note": "Completed full-suite result, replacing the previous 409-task snapshot at this URL. All 431 unique tasks validated; no unresolved infrastructure failures. Steps and tokens are raw and incomplete; token accounting has duplicate usage and the telemetry audit is unresolved. Score validation is independent. Raw traces and backend identifiers remain private.",
  "finished_at": "2026-10-17T09:07:10Z",
  "score_percent": 87.5,
  "measurement_valid": true,
  "unique_tasks": 431,
  "steps": 6321,
  "tokens": 149806005,
  "step_metric_coverage": 402,
  "token_metric_coverage": 409,
  "token_metric_status": "raw; duplicate usage and incomplete coverage; audit unresolved",
  "ordinary": {
    "passed": 368,
    "total": 391
  },
  "long_context": {
    "passed": 0,
    "total": 20,
    "timeouts": 19,
    "verifier_failures": 1
  },
  "tool_reflection": {
    "passed": 9,
    "total": 20
  },
  "passed_by_wave": [
    197,
    19,
    10,
    21,
    9,
    20,
    15,
    8,
    3,
    17,
    6,
    13,
    14,
    16,
    0,
    0,
    9
  ],
  "source_report_sha256": "48a50055546c07139fecf257f7afde65e22fdb20bcfb5bc3c12566d6045144de"
}