{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-27-008",
    "source_story_id": "tmp-story-code-attribution-style-not-self",
    "edition_id": "mp-2026-09-27-morning-0080",
    "edition_url": "https://themachinepress.com/edition/2026-09-27",
    "position": 8,
    "story_type": "dispatch",
    "section": "benchmarks-evals",
    "editorial_classification": "editorial",
    "headline": "Code Attribution Mostly Found Length, Not Identity",
    "slug": "code-attribution-mostly-found-length-not-identity",
    "dek": "Across 15 model-benchmark combinations, balanced accuracy stayed near chance and tracked solution length.",
    "summary": "Across 15 model-benchmark combinations, balanced accuracy stayed near chance and tracked solution length.",
    "body_text": "A study of model self-attribution found balanced accuracy between 49% and 58% across 15 model-benchmark combinations. Pairwise attribution scores correlated at r=0.93 with longer solutions, suggesting that style and verbosity were doing much of the work. After normalization, 10 of 12 tested conditions fell to chance and the remaining two continued to follow length, erasing the apparent self-preference in this setup.",
    "why_it_matters": "Across 15 model-benchmark combinations, balanced accuracy stayed near chance and tracked solution length.",
    "limitations": [],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-27-008/code-attribution-mostly-found-length-not-identity",
    "json_url": "https://themachinepress.com/story/mp-2026-09-27-008.json",
    "first_published_at": "2026-09-27T09:00:00.000-04:00",
    "modified_at": "2026-09-27T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-27-008-001",
        "text": "Across 15 model-benchmark combinations, balanced accuracy stayed near chance and tracked solution length.",
        "source_ids": [
          "source-2026-09-27-008"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-09-27-008"
    ],
    "tags": [
      "code generation",
      "authorship attribution",
      "evaluation bias"
    ],
    "image_url": "https://themachinepress.com/issues/2026-09-27/code-attribution-blue-screen.webp",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-27-008",
      "title": "arXiv preprint 2609.30048",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.30048",
      "canonical_url": "https://arxiv.org/abs/2609.30048",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-09-24T12:11:21.000-04:00",
      "accessed_at": "2026-09-27T08:26:41.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-27-008-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Code Attribution Mostly Found Length, Not Identity",
    "publisher": "The Machine Press",
    "published_at": "2026-09-27T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-27-008/code-attribution-mostly-found-length-not-identity"
  }
}