{
  "bucket": "blobfish-benchmark-runs-blobfish-ai-429200",
  "benchmarks": [
    {
      "id": "humaneval",
      "name": "HumanEval",
      "version": "6d43fb980f9f:full-1",
      "digest": "e7013dc4e2af46162b54486c3d2cc05538b0c8d636076b4045959ce4106c42ae",
      "format": "harbor",
      "description": "Prepared 164 of 164 public evaluation tasks (complete split). Original tests, RNG seed 0. Execution and grading are recorded separately; this package is not a measured result. Uses an agentic terminal scaffold, not the original code-only sampling protocol.",
      "tasks": 164,
      "official_tasks": 164,
      "control": false,
      "scaffolds": [
        "harbor-terminus-2",
        "harbor-codex-subscription",
        "harbor-claude-subscription",
        "harbor-deepseek"
      ]
    },
    {
      "id": "mbpp",
      "name": "MBPP",
      "version": "f46ca8374b4c:full-1",
      "digest": "70ad46ffedf5ab8e964f7792eb27024aaeb054b7d25e442f8ff0ca118591117d",
      "format": "harbor",
      "description": "Prepared 500 of 500 public evaluation tasks (complete split). Original tests, RNG seed 0. Execution and grading are recorded separately; this package is not a measured result. Uses an agentic terminal scaffold, not the original code-only sampling protocol.",
      "tasks": 500,
      "official_tasks": 500,
      "control": false,
      "scaffolds": [
        "harbor-terminus-2",
        "harbor-codex-subscription",
        "harbor-claude-subscription",
        "harbor-deepseek"
      ]
    },
    {
      "id": "devopsbench-100",
      "name": "DevOpsBench-100",
      "version": "3.2.7:revision8:full-1",
      "digest": "ec70b1bc11c0047a9b10c52556591fa61ee420b2233a03d571850041229b9631",
      "format": "harbor",
      "description": "All 100 tasks from published revision 8 (19 source-heldout, 81 source-train). Original task content, main/world images and deterministic state verifiers. Prepared full suite; results require execution.",
      "tasks": 100,
      "official_tasks": 100,
      "control": false,
      "scaffolds": [
        "harbor-terminus-2",
        "harbor-codex",
        "harbor-claude-code",
        "harbor-codex-subscription",
        "harbor-claude-subscription",
        "harbor-deepseek"
      ]
    }
  ],
  "measured": false,
  "observedAt": "2026-09-16T06:08:48.202984+00:00"
}