{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-11-06-027",
    "source_story_id": "tmp-story-coding-failure-prediction",
    "edition_id": "mp-2026-11-06-morning-0119",
    "edition_url": "https://themachinepress.com/edition/2026-11-06",
    "position": 16,
    "story_type": "dispatch",
    "section": "benchmarks-evals",
    "editorial_classification": "editorial",
    "headline": "The Failure Predictor Mostly Recognizes the Issue and Model",
    "slug": "the-failure-predictor-mostly-recognizes-the-issue-and-model",
    "dek": "On public coding-agent trajectories, issue difficulty and model identity explain far more than the partial run itself.",
    "summary": "On public coding-agent trajectories, issue difficulty and model identity explain far more than the partial run itself.",
    "body_text": "A run-independent issue-difficulty score reaches 0.92 AUROC on SWE-bench Verified trajectories, while trajectory predictors reach about 0.62. Within an issue, model identity predicts failure better than the studied predictors, and per-issue model switching provides no held-out gain. A repeated-run pilot shows that retries can rescue some failures, but current predictors cannot distinguish a bad run from a task beyond the model’s ability. The result argues for repeated-run, closed-loop evaluation rather than universal pessimism about monitoring.",
    "why_it_matters": "On public coding-agent trajectories, issue difficulty and model identity explain far more than the partial run itself.",
    "limitations": [
      "A repeated-run pilot shows that retries can rescue some failures, but current predictors cannot distinguish a bad run from a task beyond the model’s ability."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-11-06-027/the-failure-predictor-mostly-recognizes-the-issue-and-model",
    "json_url": "https://themachinepress.com/story/mp-2026-11-06-027.json",
    "first_published_at": "2026-11-06T09:00:00.000-05:00",
    "modified_at": "2026-11-06T09:00:00.000-05:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-11-06-027-001",
        "text": "On public coding-agent trajectories, issue difficulty and model identity explain far more than the partial run itself.",
        "source_ids": [
          "source-2026-11-06-016"
        ],
        "qualification": "A repeated-run pilot shows that retries can rescue some failures, but current predictors cannot distinguish a bad run from a task beyond the model’s ability."
      }
    ],
    "source_ids": [
      "source-2026-11-06-016"
    ],
    "tags": [
      "coding agents",
      "failure prediction",
      "SWE-bench"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-11-06-016",
      "title": "arXiv preprint 2611.08285, original v1",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2611.08285v1",
      "canonical_url": "https://arxiv.org/abs/2611.08285v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-11-05T11:50:11.000-05:00",
      "accessed_at": "2026-11-06T08:21:03.209-05:00",
      "supports_claim_ids": [
        "claim-mp-2026-11-06-027-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "The Failure Predictor Mostly Recognizes the Issue and Model",
    "publisher": "The Machine Press",
    "published_at": "2026-11-06T09:00:00.000-05:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-11-06-027/the-failure-predictor-mostly-recognizes-the-issue-and-model"
  }
}