{
  "schema_version": 1,
  "adapter_key": "specguard",
  "validated_at": "2026-10-20T19:09:05.676624+00:00",
  "accelerator": {
    "vendor": "NVIDIA",
    "model": "A100-SXM4-80GB"
  },
  "command": [
    "python",
    "scripts/run_specguard_component.py",
    "--checkpoint",
    "<cached-public-checkpoint>",
    "--model-id",
    "Qwen/Qwen2.5-0.5B-Instruct",
    "--revision",
    "7ae557604adf67be50417f59c2c2f167def9a775",
    "--source-commit",
    "a8063715e75bc48e731e71f64a6919dc2a939371",
    "--lean-bin",
    "<lean-4.24.0-binary>",
    "--device",
    "cuda",
    "--seeds",
    "42",
    "--max-attempts",
    "2",
    "--max-calls",
    "18",
    "--max-new-tokens",
    "1024",
    "--max-total-new-tokens",
    "8000",
    "--wall-seconds",
    "600",
    "--output",
    "<new-raw-result>"
  ],
  "dataset": {
    "id": "self-authored-increment-micro-suite",
    "public_benchmark": false,
    "revision": "18a3966acba2746dbb950bf5c05251733b0bf2ce674a849eadb4fbef5d5eb230"
  },
  "checkpoint": {
    "id": "Qwen/Qwen2.5-0.5B-Instruct",
    "revision": "7ae557604adf67be50417f59c2c2f167def9a775",
    "local_files_only": true
  },
  "seeds": [
    42
  ],
  "result": "passed",
  "result_meaning": "bounded_real_cuda_runtime_completed_not_task_or_semantic_success",
  "metrics": {
    "task_runs": 3,
    "certified_conflicts": 0,
    "formal_agreements": 0,
    "inconclusive": 3,
    "actual_model_calls": 6,
    "actual_generated_tokens": 425,
    "independent_kernel_rechecks": 0
  },
  "lean": {
    "version": "4.24.0",
    "binary_sha256": "8fadc3ed92cc9decb94c96ee9c0876e91e03a3a5e69138f32114f02789e2766c"
  },
  "diagnostic_only": true,
  "executed_model_roles": [
    "spec"
  ],
  "end_to_end_certificate_observed": false,
  "promotion_eligible": false,
  "evolve_integrated": false,
  "evaluation_tier": "l1_mechanism",
  "full_experiment_status": "not_reproduced_high_cost",
  "provenance": {
    "commit": "a8063715e75bc48e731e71f64a6919dc2a939371",
    "artifact_path": "docs/gpu-validations/specguard-a100-20261021.json",
    "executed_source_sha256": {
      "src/auto_research/agent_research/specguard.py": "3620d9170d53444b8f5e89d2783578645ad8a34f77fdc202107ebfc0529ce40e",
      "scripts/run_specguard_component.py": "88f1fc57e07bc6dc5763588f2fb5720d548ee2542611a614a4ef5bf2a0719f18"
    },
    "raw_runner_artifact_sha256": "016a3d6246c067aee653f96af9db01eebd7402b82d565459404ee050797fab48"
  }
}