[
  {
    "trial": 1,
    "session_id": "01a1c808-b550-7bc2-b431-b25698cbfb41",
    "work_started_elapsed_seconds": 21.364161249948665,
    "work_started_utc": "2026-11-11T00:36:15.839850+00:00",
    "final_validation": {
      "utc": "2026-11-11T00:39:36.784593+00:00",
      "elapsed_seconds": 222.30890516587533,
      "task_ids": [
        "task01",
        "task02",
        "task03",
        "task04",
        "task05",
        "task06",
        "task07",
        "task08",
        "task09",
        "task10"
      ],
      "returncode": 0,
      "timed_out": false,
      "seconds": 9.392628749832511
    },
    "pytest_summary": "91 passed, 397 warnings in 8.80s",
    "metadata": {
      "platform": "macOS-26.5-arm64-arm-64bit-Mach-O",
      "machine": "arm64",
      "cpu_count": 10,
      "harness_python": "3.14.5",
      "codex_version": "codex-cli 0.167.0",
      "target_python": "Python 3.14.5",
      "revision": "353c3ccdf3acdaae9acae78b5132241061b5decc",
      "evaluation_sha256": {
        "conftest.py": "9a960650d220405f2886322a473969dec5daab523bf90cf32c0f1926e092deca",
        "test_mcp_auth.py": "ac151bc4106dbc0a5eda483ae88a6c525970e4fe62d52291acb675901a43bb30",
        "test_demo_data.py": "fbd2d785b596ef8cd3eb3868d846655c7ccd032a60b2f5399c6cec1ec565ed10",
        "test_auth.py": "76e8feda4e00a7f59005aa98ffae3439dfae46c12a072773841126ff96506c78",
        "test_benchmark.py": "5b7287eb30f7bc66aa34c39f41f70fa07c39cc2bbace5df7df7365243305b9dc",
        "test_models.py": "58d165c5391f61e8eee0e9a5c0f330deee433f7f83abb992904552a65818f795",
        "test_api.py": "903671d71a524e1096a863d857f61d776d402af1f4d543e59a6ea74881b86e5b",
        "test_orders.py": "6c21755f1329b56d0bd449b128080078a2e1c7306ec48df09362a47309dcba6d"
      },
      "tasks_sha256": "efe62d068b97dd6a99fb9cf66b3addda2e6d5e7687a51ba6740d0112b132ac43",
      "harness_sha256": {
        "runner.py": "2a4b732e56896f572fbcf24fbfa7a9645e2ffdd6ecbf61f0c2f46c34dbf1cb5d",
        "__init__.py": "c4bb304adea0d61d768152331fefd4db061755b3cf4675aff854c88595e91032",
        "cli.py": "a1d7308dc883d15bc257654f5a6a97d976bed42e3c737b38c1831f35fe4522db",
        "persistent.py": "2645c9b70033e357ac33c23f2bc13e1b05dc7bea59c0374cb2ed0c80343db0bd"
      }
    }
  },
  {
    "trial": 2,
    "session_id": "01a1c80c-0409-7723-85c1-f49b1c5d57c6",
    "work_started_elapsed_seconds": 15.812654916895553,
    "work_started_utc": "2026-11-11T00:39:52.623388+00:00",
    "final_validation": {
      "utc": "2026-11-11T00:43:37.621167+00:00",
      "elapsed_seconds": 240.81043824995868,
      "task_ids": [
        "task01",
        "task02",
        "task03",
        "task04",
        "task05",
        "task06",
        "task07",
        "task08",
        "task09",
        "task10"
      ],
      "returncode": 0,
      "timed_out": false,
      "seconds": 9.24524804088287
    },
    "pytest_summary": "90 passed, 397 warnings in 8.70s",
    "metadata": {
      "platform": "macOS-26.5-arm64-arm-64bit-Mach-O",
      "machine": "arm64",
      "cpu_count": 10,
      "harness_python": "3.14.5",
      "codex_version": "codex-cli 0.167.0",
      "target_python": "Python 3.14.5",
      "revision": "353c3ccdf3acdaae9acae78b5132241061b5decc",
      "evaluation_sha256": {
        "conftest.py": "9a960650d220405f2886322a473969dec5daab523bf90cf32c0f1926e092deca",
        "test_mcp_auth.py": "ac151bc4106dbc0a5eda483ae88a6c525970e4fe62d52291acb675901a43bb30",
        "test_demo_data.py": "fbd2d785b596ef8cd3eb3868d846655c7ccd032a60b2f5399c6cec1ec565ed10",
        "test_auth.py": "76e8feda4e00a7f59005aa98ffae3439dfae46c12a072773841126ff96506c78",
        "test_benchmark.py": "5b7287eb30f7bc66aa34c39f41f70fa07c39cc2bbace5df7df7365243305b9dc",
        "test_models.py": "58d165c5391f61e8eee0e9a5c0f330deee433f7f83abb992904552a65818f795",
        "test_api.py": "903671d71a524e1096a863d857f61d776d402af1f4d543e59a6ea74881b86e5b",
        "test_orders.py": "6c21755f1329b56d0bd449b128080078a2e1c7306ec48df09362a47309dcba6d"
      },
      "tasks_sha256": "efe62d068b97dd6a99fb9cf66b3addda2e6d5e7687a51ba6740d0112b132ac43",
      "harness_sha256": {
        "runner.py": "2a4b732e56896f572fbcf24fbfa7a9645e2ffdd6ecbf61f0c2f46c34dbf1cb5d",
        "__init__.py": "c4bb304adea0d61d768152331fefd4db061755b3cf4675aff854c88595e91032",
        "cli.py": "a1d7308dc883d15bc257654f5a6a97d976bed42e3c737b38c1831f35fe4522db",
        "persistent.py": "2645c9b70033e357ac33c23f2bc13e1b05dc7bea59c0374cb2ed0c80343db0bd"
      }
    }
  },
  {
    "trial": 3,
    "session_id": "01a1c80f-ae80-7b53-b787-eba4d05730ae",
    "work_started_elapsed_seconds": 15.198350459104404,
    "work_started_utc": "2026-11-11T00:43:52.841679+00:00",
    "final_validation": {
      "utc": "2026-11-11T00:47:28.254817+00:00",
      "elapsed_seconds": 230.6114907921292,
      "task_ids": [
        "task01",
        "task02",
        "task03",
        "task04",
        "task05",
        "task06",
        "task07",
        "task08",
        "task09",
        "task10"
      ],
      "returncode": 0,
      "timed_out": false,
      "seconds": 9.678058875026181
    },
    "pytest_summary": "93 passed, 397 warnings in 9.08s",
    "metadata": {
      "platform": "macOS-26.5-arm64-arm-64bit-Mach-O",
      "machine": "arm64",
      "cpu_count": 10,
      "harness_python": "3.14.5",
      "codex_version": "codex-cli 0.167.0",
      "target_python": "Python 3.14.5",
      "revision": "353c3ccdf3acdaae9acae78b5132241061b5decc",
      "evaluation_sha256": {
        "conftest.py": "9a960650d220405f2886322a473969dec5daab523bf90cf32c0f1926e092deca",
        "test_mcp_auth.py": "ac151bc4106dbc0a5eda483ae88a6c525970e4fe62d52291acb675901a43bb30",
        "test_demo_data.py": "fbd2d785b596ef8cd3eb3868d846655c7ccd032a60b2f5399c6cec1ec565ed10",
        "test_auth.py": "76e8feda4e00a7f59005aa98ffae3439dfae46c12a072773841126ff96506c78",
        "test_benchmark.py": "5b7287eb30f7bc66aa34c39f41f70fa07c39cc2bbace5df7df7365243305b9dc",
        "test_models.py": "58d165c5391f61e8eee0e9a5c0f330deee433f7f83abb992904552a65818f795",
        "test_api.py": "903671d71a524e1096a863d857f61d776d402af1f4d543e59a6ea74881b86e5b",
        "test_orders.py": "6c21755f1329b56d0bd449b128080078a2e1c7306ec48df09362a47309dcba6d"
      },
      "tasks_sha256": "efe62d068b97dd6a99fb9cf66b3addda2e6d5e7687a51ba6740d0112b132ac43",
      "harness_sha256": {
        "runner.py": "2a4b732e56896f572fbcf24fbfa7a9645e2ffdd6ecbf61f0c2f46c34dbf1cb5d",
        "__init__.py": "c4bb304adea0d61d768152331fefd4db061755b3cf4675aff854c88595e91032",
        "cli.py": "a1d7308dc883d15bc257654f5a6a97d976bed42e3c737b38c1831f35fe4522db",
        "persistent.py": "2645c9b70033e357ac33c23f2bc13e1b05dc7bea59c0374cb2ed0c80343db0bd"
      }
    }
  }
]
