{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-06-001",
    "source_story_id": "tmp-lead-reasoning-step-importance",
    "edition_id": "mp-2026-09-06-morning-0059",
    "edition_url": "https://themachinepress.com/edition/2026-09-06",
    "position": 1,
    "story_type": "lead",
    "section": "safety",
    "editorial_classification": "editorial",
    "headline": "The Reasoning Was Readable. Its Important Steps Were Still Hidden",
    "slug": "the-reasoning-was-readable-its-important-steps-were-still-hidden",
    "dek": "Rollout-based advantage tests found that language-model judges could identify consequential chain-of-thought steps better than chance—but far below the experiment's noise ceiling.",
    "summary": "Rollout-based advantage tests found that language-model judges could identify consequential chain-of-thought steps better than chance—but far below the experiment's noise ceiling.",
    "body_text": "Researchers estimated each reasoning step's functional importance by measuring how much including it changed the expected final reward across Monte Carlo continuations. Capable language models beat a prevalence baseline when asked to identify high-advantage steps from the text alone, yet remained well short of the noise ceiling. Fine-tuning a step critic helped more on wrong answers than correct ones. The result does not show that reasoning text is useless; it shows that readable prose only partially reveals which step actually carries the answer. That distinction matters when traces are used for error diagnosis, process rewards or claims of interpretability.",
    "why_it_matters": "Rollout-based advantage tests found that language-model judges could identify consequential chain-of-thought steps better than chance—but far below the experiment's noise ceiling.",
    "limitations": [
      "The result does not show that reasoning text is useless; it shows that readable prose only partially reveals which step actually carries the answer."
    ],
    "importance": 10,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-06-001/the-reasoning-was-readable-its-important-steps-were-still-hidden",
    "json_url": "https://themachinepress.com/story/mp-2026-09-06-001.json",
    "first_published_at": "2026-09-06T09:00:00.000-04:00",
    "modified_at": "2026-09-06T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-06-001-001",
        "text": "Rollout-based advantage tests found that language-model judges could identify consequential chain-of-thought steps better than chance—but far below the experiment's noise ceiling.",
        "source_ids": [
          "source-2026-09-06-001"
        ],
        "qualification": "The result does not show that reasoning text is useless; it shows that readable prose only partially reveals which step actually carries the answer."
      }
    ],
    "source_ids": [
      "source-2026-09-06-001"
    ],
    "tags": [
      "chain of thought",
      "interpretability",
      "process evaluation"
    ],
    "image_url": "https://themachinepress.com/issues/2026-09-06/lead-reasoning-importance.png",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-06-001",
      "title": "arXiv preprint 2609.04194",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.04194",
      "canonical_url": "https://arxiv.org/abs/2609.04194",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-09-03T13:59:08.000-04:00",
      "accessed_at": "2026-09-06T08:20:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-06-001-001"
      ]
    }
  ],
  "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 Reasoning Was Readable. Its Important Steps Were Still Hidden",
    "publisher": "The Machine Press",
    "published_at": "2026-09-06T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-06-001/the-reasoning-was-readable-its-important-steps-were-still-hidden"
  }
}