{
  "container_limits": {
    "mem_limit": 8589934592,
    "memswap_limit": 8589934592,
    "nano_cpus": 4000000000,
    "network_mode": "none",
    "pids_limit": 512
  },
  "created_at": "2026-10-13T09:56:56.171475Z",
  "dataset": "princeton-nlp/SWE-bench_Verified",
  "dataset_snapshot": {
    "files": {
      "test-00000-of-00001.parquet": "a45b1fe4e2f0c8390b2b2938ac83e92ed5979000856808f3679c07812e9e6dcd"
    },
    "revision": "c104f840cc67f8b6eec6f759ebc8b2693d585d4a",
    "root": "<redacted-local-dataset-root>"
  },
  "endpoint": {
    "base_url": "<redacted-local-endpoint>",
    "model": "glm53-flash-exl3-4bpw-v84-fp8-tp2-dcp1-c4-327k-r11-vision8-apc-nospec"
  },
  "failure": null,
  "harnesses": {
    "agent": {
      "name": "mini-swe-agent",
      "revision": "a83fcae82d2a08f0ee0c688f9d137b3566c097f8"
    },
    "grader": {
      "name": "swe-bench",
      "revision": "f7bbbb2ccdf479001d6467c9e34af59e44a840f9"
    },
    "grader_containment": {
      "adapter_sha256": "23c56b27d62b79c39aca6dcfdd4c504030e241e3a6941620cdc6e947b9035a7b",
      "receipt": "<redacted-local-containment-receipt>",
      "source_hashes": {
        "swebench/harness/docker_build.py": "5278842b60a7d38256f95f93c915dc84de2b8b4f286e9baae1b19280f768e484",
        "swebench/harness/run_evaluation.py": "6959f0b4e4eaf979771f529b88e3e9df1daa7fe86bc4291feec2e7d320bf7f2e"
      }
    },
    "python_environment": {
      "architecture": "x86_64",
      "platform": "linux",
      "python": {
        "implementation": "CPython",
        "version": "3.14.4"
      },
      "resolved_packages_sha256": "809336a2a74041fa07adad60eaca2208acd936f3728d2368d22f98657576a37a",
      "schema": "anvil-serving.swe-python-environment/v1"
    }
  },
  "image_observations": [],
  "instances": [
    {
      "duration_s": null,
      "exit_status": "Submitted",
      "failure_class": null,
      "grader": {
        "completed": true,
        "name": "swe-bench",
        "report_sha256": "d0f2ff7e24f6e1b02d08b0e68b71ded24e917f84b45636702b6215526c24d656",
        "resolved": true,
        "revision": "f7bbbb2ccdf479001d6467c9e34af59e44a840f9"
      },
      "instance_id": "sphinx-doc__sphinx-9658",
      "prediction_sha256": "6a492d8268228afd84ff364160bde7d0d99f2a9e8c483333d8e0459ec7897f46",
      "prompt_identity": {
        "dataset": "princeton-nlp/SWE-bench_Verified",
        "split": "test"
      },
      "request_id_count": 28,
      "request_ids": "<redacted-request-ids>",
      "tokens": {
        "completion_tokens": null,
        "prompt_tokens": null,
        "total_tokens": null
      },
      "trajectory": {
        "path": "<redacted-local-trajectory>",
        "sha256": "db58af68b280afa93a70d11aa12fb9af3021bbde471e2d78b970b79c249e51d0"
      }
    },
    {
      "duration_s": null,
      "exit_status": "Submitted",
      "failure_class": null,
      "grader": {
        "completed": true,
        "name": "swe-bench",
        "report_sha256": "d0f2ff7e24f6e1b02d08b0e68b71ded24e917f84b45636702b6215526c24d656",
        "resolved": false,
        "revision": "f7bbbb2ccdf479001d6467c9e34af59e44a840f9"
      },
      "instance_id": "sympy__sympy-12096",
      "prediction_sha256": "3d8af9d53e3564bfda053a3b8be6f4c4e4856ba905bee5d62beebc9c9d9a46b5",
      "prompt_identity": {
        "dataset": "princeton-nlp/SWE-bench_Verified",
        "split": "test"
      },
      "request_id_count": 57,
      "request_ids": "<redacted-request-ids>",
      "tokens": {
        "completion_tokens": null,
        "prompt_tokens": null,
        "total_tokens": null
      },
      "trajectory": {
        "path": "<redacted-local-trajectory>",
        "sha256": "5226a0f656b0e0e99aa518fed809b16216dfb21ebbb51a45fd3fe73b77894d2c"
      }
    },
    {
      "duration_s": null,
      "exit_status": "Submitted",
      "failure_class": null,
      "grader": {
        "completed": true,
        "name": "swe-bench",
        "report_sha256": "d0f2ff7e24f6e1b02d08b0e68b71ded24e917f84b45636702b6215526c24d656",
        "resolved": false,
        "revision": "f7bbbb2ccdf479001d6467c9e34af59e44a840f9"
      },
      "instance_id": "django__django-13512",
      "prediction_sha256": "40eb3bf9deb66aa85579f2be5ab074a4aa7b697a3fbbe9e826f3109c6cdaaa26",
      "prompt_identity": {
        "dataset": "princeton-nlp/SWE-bench_Verified",
        "split": "test"
      },
      "request_id_count": 43,
      "request_ids": "<redacted-request-ids>",
      "tokens": {
        "completion_tokens": null,
        "prompt_tokens": null,
        "total_tokens": null
      },
      "trajectory": {
        "path": "<redacted-local-trajectory>",
        "sha256": "becebafc7043fdb3a032ed513c4fc561e3d56361b930f2eefd698a01ea1fa955"
      }
    },
    {
      "duration_s": null,
      "exit_status": "Submitted",
      "failure_class": null,
      "grader": {
        "completed": true,
        "name": "swe-bench",
        "report_sha256": "d0f2ff7e24f6e1b02d08b0e68b71ded24e917f84b45636702b6215526c24d656",
        "resolved": true,
        "revision": "f7bbbb2ccdf479001d6467c9e34af59e44a840f9"
      },
      "instance_id": "pydata__xarray-4356",
      "prediction_sha256": "35306b4c7f2233fc00b52ac637bbd9f32ceb4a533d37ce1d1780601d325a3cf9",
      "prompt_identity": {
        "dataset": "princeton-nlp/SWE-bench_Verified",
        "split": "test"
      },
      "request_id_count": 36,
      "request_ids": "<redacted-request-ids>",
      "tokens": {
        "completion_tokens": null,
        "prompt_tokens": null,
        "total_tokens": null
      },
      "trajectory": {
        "path": "<redacted-local-trajectory>",
        "sha256": "e436f6e32a44754dc2cf350d5431937a8e122ec91eee11dd7fe191d88da05d5a"
      }
    },
    {
      "duration_s": null,
      "exit_status": "Submitted",
      "failure_class": null,
      "grader": {
        "completed": true,
        "name": "swe-bench",
        "report_sha256": "d0f2ff7e24f6e1b02d08b0e68b71ded24e917f84b45636702b6215526c24d656",
        "resolved": false,
        "revision": "f7bbbb2ccdf479001d6467c9e34af59e44a840f9"
      },
      "instance_id": "pylint-dev__pylint-8898",
      "prediction_sha256": "91caf6e9855e95f7358f3767a8cc182a56266d97b6ebd3eb4813d0697616bd3c",
      "prompt_identity": {
        "dataset": "princeton-nlp/SWE-bench_Verified",
        "split": "test"
      },
      "request_id_count": 32,
      "request_ids": "<redacted-request-ids>",
      "tokens": {
        "completion_tokens": null,
        "prompt_tokens": null,
        "total_tokens": null
      },
      "trajectory": {
        "path": "<redacted-local-trajectory>",
        "sha256": "eba00e55f39c5a174e761569ea01781425d5d2443eb6436c72d4401b2ae88da6"
      }
    }
  ],
  "official_grader_complete": true,
  "ownership_id": "qualification-20261013",
  "paired_image_ids": null,
  "plan_sha256": "a7f471fadd901d85c7f84e4c6b1d0fdfa90729ea199b619c6f022a1ec2846bd2",
  "profile": "scout",
  "profile_sha256": "fce0e6e1e7497585bdfc6f76507d3baaff99a6b4f174d841d268de20c68de7f6",
  "promotion": {
    "authorized": false,
    "message": "Benchmark evidence does not authorize model promotion."
  },
  "request_controls": {
    "reasoning_effort": null,
    "sampling": {
      "temperature": {
        "effective_request": 0.6,
        "requested": 0.6,
        "sent": true
      },
      "top_p": {
        "effective_request": 0.95,
        "requested": 0.95,
        "sent": true
      }
    },
    "thinking_mode": "enabled"
  },
  "run_id": "qualification-20261013-baseline-swe5",
  "schema": "anvil-serving.swe-run/v1",
  "selection": {
    "instance_ids": [
      "sphinx-doc__sphinx-9658",
      "sympy__sympy-12096",
      "django__django-13512",
      "pydata__xarray-4356",
      "pylint-dev__pylint-8898"
    ],
    "kind": "explicit_instance_ids",
    "sha256": "622934396fb97d950f803dfbb0791f6a398a5a6fe0fd94cc27b329f1a953cac4"
  },
  "split": "test",
  "stage_count": 2,
  "stages": "<redacted-runtime-stage-transcripts>",
  "state": "completed",
  "summary": {
    "attempted": 5,
    "graded": 5,
    "resolve_rate": 0.4,
    "resolved": 2
  },
  "task_container_network": "none"
}