{
  "arms": [
    {
      "arm": "A",
      "combined_log_sha256": "30c11c13f03727dd7c9fb11dd45c83b0c0314b53418782d2154c906ebe2b6a81",
      "duration_ms": 42083,
      "error": null,
      "exit_code": 0,
      "final_message_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
      "label": "One sentence",
      "reported_tokens": 40329,
      "score": {
        "answer_correct": 10,
        "answer_file": true,
        "answer_sha256": "668fe5e5d725542556cfbfdc1eb6cddbf2d61d398b360992e8242112bf79e573",
        "claim_status_correct": 9,
        "materialization": 1,
        "parseable": true,
        "schema_exact": true,
        "severe_false_support": [],
        "t5_sha256": "08cfbc3be7d2a5bb21ebd97b9a986db15adaa2c85e019abae6729e0999292204",
        "task_detail": {
          "T1": {
            "answer_ok": true,
            "claim_status_ok": true,
            "point": 1
          },
          "T10": {
            "answer_ok": true,
            "claim_status_ok": true,
            "point": 1
          },
          "T2": {
            "answer_ok": true,
            "claim_status_ok": true,
            "point": 1
          },
          "T3": {
            "answer_ok": true,
            "claim_status_ok": true,
            "point": 1
          },
          "T4": {
            "answer_ok": true,
            "claim_status_ok": true,
            "point": 1
          },
          "T5": {
            "answer_ok": true,
            "claim_status_ok": true,
            "point": 1
          },
          "T6": {
            "answer_ok": true,
            "claim_status_ok": true,
            "point": 1
          },
          "T7": {
            "answer_ok": true,
            "claim_status_ok": true,
            "point": 1
          },
          "T8": {
            "answer_ok": true,
            "claim_status_ok": true,
            "point": 1
          },
          "T9": {
            "answer_ok": true,
            "claim_status_ok": false,
            "point": 0
          }
        },
        "task_points": 9,
        "total_points": 10
      },
      "status": "completed"
    }
  ],
  "benchmark_id": "PARS-PILOT-2026-08-16-V2",
  "claim_boundary": [
    "A self-run receipt is replication input, not canonical audit evidence.",
    "The score applies only to the frozen deterministic task bank.",
    "The receipt does not establish sampling equivalence, PARS superiority, or an independent semantic judgment."
  ],
  "codex": {
    "authentication": "ChatGPT-authenticated Codex reported by login status",
    "codex_version": "codex-cli 0.148.0-alpha.9"
  },
  "completed_at": "2026-08-16T10:57:29.744208Z",
  "created_at": "2026-08-16T10:56:47.641920Z",
  "environment": {
    "api_key_variables_removed": [
      "ANTHROPIC_API_KEY",
      "AZURE_OPENAI_API_KEY",
      "CODEX_API_KEY",
      "GEMINI_API_KEY",
      "GOOGLE_API_KEY",
      "OPENAI_API_KEY",
      "XAI_API_KEY"
    ],
    "platform": "macOS-26.5.2-arm64-arm-64bit",
    "python": "3.9.6"
  },
  "hashes": {
    "answer_key_sha256": "cb4bde32b502eceaba5490567a1bd0d86ef69fa71cb23bcfc8b8941cd09fe88d",
    "conditions_sha256": "318b3217474001be9ec6275f53f3c817845cc6281dd6dc2985f5be094ff184af",
    "pars_skill_sha256": "424bf5359795b899126f5ce0c8bf6cc99fb2675b800f1ab915b410dce240b602",
    "task_packet_sha256": "570343446e66c5f80e6127595f9f355ddedf54152c57367dd99a55e3f3a6fadd"
  },
  "order": [
    "A"
  ],
  "status": "complete",
  "upstream_commit": "57ac9970fa324c7c2a6f613fe32f5cc0577d3c3e"
}