{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-09-015",
    "source_story_id": "tmp-brief-ontrack-streaming-agent-monitor",
    "edition_id": "mp-2026-10-09-morning-0091",
    "edition_url": "https://themachinepress.com/edition/2026-10-09",
    "position": 17,
    "story_type": "ticker",
    "section": "developer-tools",
    "editorial_classification": "editorial",
    "headline": "Eight Steps Were Enough to Flag Some Failing Runs",
    "slug": "eight-steps-were-enough-to-flag-some-failing-runs",
    "dek": "OnTrack compares streaming agent structure with successful trajectories at about a millisecond per step.",
    "summary": "OnTrack compares streaming agent structure with successful trajectories at about a millisecond per step.",
    "body_text": "On SWE-bench trajectories, the first eight steps improved failure ranking by 0.057 AUROC over content-similarity methods. A tested abort policy saved about 18% of compute destined for failing runs; five of six aborts were correct. Monitoring weakened as access to reference runs and schemas decreased.",
    "why_it_matters": "OnTrack compares streaming agent structure with successful trajectories at about a millisecond per step.",
    "limitations": [],
    "importance": 7,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-015/eight-steps-were-enough-to-flag-some-failing-runs",
    "json_url": "https://themachinepress.com/story/mp-2026-10-09-015.json",
    "first_published_at": "2026-10-09T09:00:00.000-04:00",
    "modified_at": "2026-10-10T03:43:44.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-09-015-001",
        "text": "OnTrack compares streaming agent structure with successful trajectories at about a millisecond per step.",
        "source_ids": [
          "source-2026-10-09-017"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-10-09-017"
    ],
    "tags": [
      "agent monitoring",
      "early intervention"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-09-017",
      "title": "arXiv preprint 2610.12375, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12375v1",
      "canonical_url": "https://arxiv.org/abs/2610.12375v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:32:08.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-015-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Eight Steps Were Enough to Flag Some Failing Runs",
    "publisher": "The Machine Press",
    "published_at": "2026-10-09T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-015/eight-steps-were-enough-to-flag-some-failing-runs"
  }
}