{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-21-027",
    "source_story_id": "tmp-story-composed-task-budget",
    "edition_id": "mp-2026-10-21-morning-0103",
    "edition_url": "https://themachinepress.com/edition/2026-10-21",
    "position": 16,
    "story_type": "dispatch",
    "section": "benchmarks-evals",
    "editorial_classification": "editorial",
    "headline": "Put More Tasks in the Prompt, and Each One Gets Less Thought",
    "slug": "put-more-tasks-in-the-prompt-and-each-one-gets-less-thought",
    "dek": "Errors under composition tracked an implicit per-task reasoning budget even when the full prompt fit comfortably in context.",
    "summary": "Errors under composition tracked an implicit per-task reasoning budget even when the full prompt fit comfortably in context.",
    "body_text": "Models that solve tasks separately can fail when those same tasks are bundled. Across open-weight systems and Claude Opus 5, the authors found degradation beyond independent error accumulation: models used fewer tokens per task under composition. Problems needing longer traces alone were more fragile when grouped. Sorting tasks by fragility manipulated that allocation, cutting token use by 43 percent across tested math and language tasks while preserving accuracy. The mechanism is inferred from behavioral measurements and may not transfer to every model or prompting setup.",
    "why_it_matters": "Errors under composition tracked an implicit per-task reasoning budget even when the full prompt fit comfortably in context.",
    "limitations": [
      "The mechanism is inferred from behavioral measurements and may not transfer to every model or prompting setup."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-21-027/put-more-tasks-in-the-prompt-and-each-one-gets-less-thought",
    "json_url": "https://themachinepress.com/story/mp-2026-10-21-027.json",
    "first_published_at": "2026-10-21T09:00:00.000-04:00",
    "modified_at": "2026-10-21T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-21-027-001",
        "text": "Errors under composition tracked an implicit per-task reasoning budget even when the full prompt fit comfortably in context.",
        "source_ids": [
          "source-2026-10-21-016"
        ],
        "qualification": "The mechanism is inferred from behavioral measurements and may not transfer to every model or prompting setup."
      }
    ],
    "source_ids": [
      "source-2026-10-21-016"
    ],
    "tags": [
      "language models",
      "reasoning",
      "task composition"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-21-016",
      "title": "arXiv preprint 2610.27190, original v1",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2610.27190v1",
      "canonical_url": "https://arxiv.org/abs/2610.27190v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-20T13:59:00.000-04:00",
      "accessed_at": "2026-10-21T08:27:23.757-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-21-027-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Put More Tasks in the Prompt, and Each One Gets Less Thought",
    "publisher": "The Machine Press",
    "published_at": "2026-10-21T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-21-027/put-more-tasks-in-the-prompt-and-each-one-gets-less-thought"
  }
}