{
  "created_at": "2026-10-26T18:02:46.588234Z",
  "failure": {
    "class": "resource_exhaustion",
    "code": "budget_exhausted",
    "message": "benchmark suite did not complete"
  },
  "partial": true,
  "provenance": {
    "endpoint": {
      "auth_env": "ANVIL_EVALUATION_TOKEN",
      "base_url": "[REDACTED_PRIVATE_VALUE]",
      "model": "llm.evaluation.flare"
    },
    "finished_at": "2026-10-26T18:02:46.588074Z",
    "started_at": "2026-10-26T17:09:25.179216Z",
    "submitted_at": "2026-10-26T16:35:11+00:00",
    "worker": {
      "id": "Primary Node"
    }
  },
  "results": {
    "evidence": {
      "completeness": "failed",
      "created_at": "2026-10-26T18:02:46.578222Z",
      "evidence_kind": "measured",
      "failure": {
        "class": "resource_exhaustion",
        "code": "budget_exhausted",
        "message": "benchmark suite did not complete"
      },
      "identities": {
        "concurrency": {
          "requests": 1,
          "workers": 1
        },
        "context": {
          "advertised_context_tokens": 262144,
          "effective_max_model_len": 262144,
          "observation": "native /v1/models max_model_len"
        },
        "dataset": {
          "name": "SWE-bench_Verified",
          "split": "test"
        },
        "hardware": {
          "gpu_name": "NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition",
          "gpu_uuids": [
            "[REDACTED_PUBLIC_HARDWARE_ID]"
          ]
        },
        "harnesses": {
          "mini-swe-agent": {
            "revision": "a83fcae82d2a08f0ee0c688f9d137b3566c097f8",
            "source": "https://github.com/SWE-agent/mini-swe-agent"
          },
          "swe-bench": {
            "revision": "f7bbbb2ccdf479001d6467c9e34af59e44a840f9",
            "source": "https://github.com/SWE-bench/SWE-bench"
          },
          "swe-bench-verified": {
            "revision": "c104f840cc67f8b6eec6f759ebc8b2693d585d4a",
            "source": "https://huggingface.co/datasets/princeton-nlp/SWE-bench_Verified"
          },
          "worker-base": {
            "image": "python:3.12-slim@sha256:57cd7c3a7a273101a6485ba99423ee568157882804b1124b4dd04266317710de"
          }
        },
        "image": {
          "id": "sha256:67124437a73011d63c2b25c939b9c48581eba90e51c6e565d4c808d161d10fc6",
          "reference": "ghcr.io/local-inference-lab/flare:release-2193455-k7-pf4096-20261022@sha256:67124437a73011d63c2b25c939b9c48581eba90e51c6e565d4c808d161d10fc6"
        },
        "model": {
          "repository": "local-inference-lab/Qwen3.8-Flash-Next-FLARE",
          "revision": "431b302155d53e296aa51633a997b4b5fd92e751"
        },
        "runtime": {
          "benchmark_source_revision": "a5ccb7454ad49f47e5a953b130696fff69cdb829",
          "engine": "flare-2193455651794642",
          "operator_provenance": {
            "source_commit": "c87dc0ac4f5399b3cb5da17d9cf8babafaf2e143",
            "source_tree": "4e35fc3ecd99d97abf8fd3a297fe7583f7c51b53",
            "tree_identical_to_main_commit": "d6997162330f41593663fbf9f9ebe0618ebb9ad8"
          },
          "recipe_sha256": "6fd91446e710888c9bf9a71a2a52b2d38ba65634110134f02398c3143c2513fa",
          "request_overrides": {
            "temperature": 0.6,
            "top_p": 0.95
          },
          "sampling_defaults_observed": {
            "temperature": 1.0,
            "top_k": 20,
            "top_p": 0.949999988079071
          },
          "source_revision": "2193455651794642debdaad0dda91e77ab675031",
          "stored_data_verification": {
            "evidence": "startup reported Stored data: Not checked (--verify-weights to check)",
            "full_payload_verified": false,
            "status": "not_checked"
          },
          "unspecified_sampling_controls_equivalence": "not_asserted"
        },
        "served_model": {
          "configured_context_tokens": 262144,
          "id": "qwen38-flash-next-flare-tp1-c16-mtp5",
          "router_alias": "llm.evaluation.flare"
        },
        "topology": {
          "client_topology": "co-resident",
          "gpu_uuids": [
            "[REDACTED_PUBLIC_HARDWARE_ID]"
          ],
          "max_concurrency": 16,
          "model_host_id": "Primary Node",
          "mtp_draft_tokens": 5,
          "tensor_parallel": 1
        }
      },
      "promotion": {
        "authorized": false,
        "message": "Benchmark evidence does not authorize model promotion; promotion is a separate human decision."
      },
      "run": {
        "ownership_id": "evaluation-platform",
        "profile": "scout",
        "run_id": "flare-r2-swe-scout-r1-20261026",
        "spec_sha256": "51b151c0eaa7329eff862b925333202975f086e2abe596ae442d36ab3e0540ff",
        "suite": "swe"
      },
      "schema": "anvil-serving.benchmark-evidence/v1",
      "stages": [
        {
          "evidence": [
            {
              "bytes": 11054,
              "path": "[REDACTED_PRIVATE_VALUE]",
              "sha256": "1f927c5d7a498b27776f8027d3d88da728bca3c841842f23cda82e8ea0764e74"
            }
          ],
          "name": "asset_preparation",
          "sequence": 0,
          "status": "completed"
        },
        {
          "evidence": [
            {
              "bytes": 2207,
              "path": "[REDACTED_PRIVATE_VALUE]",
              "sha256": "e85d98b97306dd9b33951b8e00f23f81ffd77b10a959f2c7aa90f667defb77b0"
            }
          ],
          "name": "preflight",
          "sequence": 1,
          "status": "completed"
        },
        {
          "evidence": [
            {
              "bytes": 4425,
              "path": "[REDACTED_PRIVATE_VALUE]",
              "sha256": "46b947bcf9f0d00e4d4b710925010fb3eb5d9824d86413e20c1ccfdc92e86329"
            }
          ],
          "name": "swe_canonical_retention",
          "sequence": 2,
          "status": "completed"
        },
        {
          "evidence": [
            {
              "bytes": 731874,
              "path": "[REDACTED_PRIVATE_VALUE]",
              "sha256": "c87e71341e9573c39a9fdd6faa8d1703ca889fac0a1d7c4e007cf87f7572f46c"
            }
          ],
          "name": "swe",
          "sequence": 3,
          "status": "failed"
        }
      ],
      "summary": {
        "attempted": true,
        "completed": false
      }
    }
  },
  "run": {
    "ownership_id": "evaluation-platform",
    "profile": "scout",
    "run_id": "flare-r2-swe-scout-r1-20261026",
    "spec_sha256": "51b151c0eaa7329eff862b925333202975f086e2abe596ae442d36ab3e0540ff",
    "suite": "swe"
  },
  "schema": "anvil-serving.benchmark-result/v1",
  "status": "failed"
}