{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-11-12-005",
    "source_story_id": "tmp-story-agent-errors-enter-early",
    "edition_id": "mp-2026-11-12-morning-0125",
    "edition_url": "https://themachinepress.com/edition/2026-11-12",
    "position": 5,
    "story_type": "dispatch",
    "section": "safety-security",
    "editorial_classification": "editorial",
    "headline": "The Agent's Failure Usually Starts Near the Beginning",
    "slug": "the-agent-s-failure-usually-starts-near-the-beginning",
    "dek": "A trajectory audit finds contradictions were already present before failure in most reviewed runs, but the systems rarely acted on them.",
    "summary": "A trajectory audit finds contradictions were already present before failure in most reviewed runs, but the systems rarely acted on them.",
    "body_text": "Researchers read 95 failed multi-agent trajectories in full and annotated where each fault originated, how it propagated, what evidence contradicted it and which control could have reached it. The median fault entered at 18% of the run, usually as a wrong task-state decision or fabrication. In 72% of failures the trajectory already contained evidence against the error; in 88% of those cases no agent acted on it. The audit also found that 45% of one corpus's hand-crafted failures ended in a provider content-filter crash followed by a harness-generated answer, exposing a failure class that responsible-agent labels miss.",
    "why_it_matters": "A trajectory audit finds contradictions were already present before failure in most reviewed runs, but the systems rarely acted on them.",
    "limitations": [],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-11-12-005/the-agent-s-failure-usually-starts-near-the-beginning",
    "json_url": "https://themachinepress.com/story/mp-2026-11-12-005.json",
    "first_published_at": "2026-11-12T09:00:00.000-05:00",
    "modified_at": "2026-11-12T09:00:00.000-05:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-11-12-005-001",
        "text": "A trajectory audit finds contradictions were already present before failure in most reviewed runs, but the systems rarely acted on them.",
        "source_ids": [
          "source-2026-11-12-005"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-11-12-005"
    ],
    "tags": [
      "agents",
      "reliability",
      "trajectories",
      "controls"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-11-12-005",
      "title": "arXiv:2611.13254v1",
      "publisher": "arXiv / Cornell University",
      "url": "https://arxiv.org/abs/2611.13254v1",
      "canonical_url": "https://arxiv.org/abs/2611.13254v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-11-09T16:25:08.000-05:00",
      "accessed_at": "2026-11-12T08:24:40.465-05:00",
      "supports_claim_ids": [
        "claim-mp-2026-11-12-005-001"
      ]
    }
  ],
  "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 Agent's Failure Usually Starts Near the Beginning",
    "publisher": "The Machine Press",
    "published_at": "2026-11-12T09:00:00.000-05:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-11-12-005/the-agent-s-failure-usually-starts-near-the-beginning"
  }
}