{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-01-005",
    "source_story_id": "tmp-story-readout-gap-reasoning",
    "edition_id": "mp-2026-09-01-morning-0054",
    "edition_url": "https://themachinepress.com/edition/2026-09-01",
    "position": 5,
    "story_type": "dispatch",
    "section": "benchmarks-evals",
    "editorial_classification": "editorial",
    "headline": "The Correct Answer Survived Behind the Wrong Score",
    "slug": "the-correct-answer-survived-behind-the-wrong-score",
    "dek": "A two-parameter, label-free correction recovered decisions from hidden states after native sequence scoring collapsed.",
    "summary": "A two-parameter, label-free correction recovered decisions from hidden states after native sequence scoring collapsed.",
    "body_text": "The study distinguishes absent reasoning from an output bottleneck: hidden-state probes decoded correct answers even when native sequence scores failed under structural biases. A minimal additive correction fitted on as few as 25 unlabeled examples recovered 9 to 34 accuracy points for Qwen3.5 models and transferred to OLMo-2-1B and Llama-3.1-8B. Hard-instance and permutation controls supported an instance-specific signal, but the result narrows how some benchmark failures should be interpreted rather than proving latent correctness in every error.",
    "why_it_matters": "A two-parameter, label-free correction recovered decisions from hidden states after native sequence scoring collapsed.",
    "limitations": [],
    "importance": 9,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-01-005/the-correct-answer-survived-behind-the-wrong-score",
    "json_url": "https://themachinepress.com/story/mp-2026-09-01-005.json",
    "first_published_at": "2026-09-01T09:00:00.000-04:00",
    "modified_at": "2026-09-01T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-01-005-001",
        "text": "A two-parameter, label-free correction recovered decisions from hidden states after native sequence scoring collapsed.",
        "source_ids": [
          "source-2026-09-01-005"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-09-01-005"
    ],
    "tags": [
      "reasoning",
      "hidden states",
      "calibration"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-01-005",
      "title": "arXiv preprint 2608.31068",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2608.31068",
      "canonical_url": "https://arxiv.org/abs/2608.31068",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-08-31T12:43:55.000-04:00",
      "accessed_at": "2026-09-01T08:25:35.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-01-005-001"
      ]
    }
  ],
  "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 Correct Answer Survived Behind the Wrong Score",
    "publisher": "The Machine Press",
    "published_at": "2026-09-01T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-01-005/the-correct-answer-survived-behind-the-wrong-score"
  }
}