{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-08-10-014",
    "source_story_id": "tmp-story-fisher-r1-hypothesis-testing",
    "edition_id": "mp-2026-08-10-morning-0032",
    "edition_url": "https://themachinepress.com/edition/2026-08-10",
    "position": 14,
    "story_type": "dispatch",
    "section": "benchmarks-evals",
    "editorial_classification": "editorial",
    "headline": "The P-Value Passed. The Assumption Did Not",
    "slug": "the-p-value-passed-the-assumption-did-not",
    "dek": "P-Bench targets statistical mistakes that survive correctly executed code but invalidate an agent's scientific conclusion.",
    "summary": "P-Bench targets statistical mistakes that survive correctly executed code but invalidate an agent's scientific conclusion.",
    "body_text": "P-Bench contains 425 open-ended hypothesis-testing tasks across economics, biology and medicine, requiring an agent to choose a method, compute a p-value and interpret the result. The authors say current agents often execute code correctly while violating assumptions behind the test. Their open-weight Fisher-R1-14B model, trained with verified statistical rewards, improved single-trial success by 21 percent relative to DeepSeek-V4-Pro on average. The benchmark and model are preprint contributions, not validation for unsupervised science.",
    "why_it_matters": "P-Bench targets statistical mistakes that survive correctly executed code but invalidate an agent's scientific conclusion.",
    "limitations": [
      "The benchmark and model are preprint contributions, not validation for unsupervised science."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-10-014/the-p-value-passed-the-assumption-did-not",
    "json_url": "https://themachinepress.com/story/mp-2026-08-10-014.json",
    "first_published_at": "2026-08-10T09:00:00.000-04:00",
    "modified_at": "2026-08-10T15:50:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-08-10-014-001",
        "text": "P-Bench targets statistical mistakes that survive correctly executed code but invalidate an agent's scientific conclusion.",
        "source_ids": [
          "source-2026-08-10-014"
        ],
        "qualification": "The benchmark and model are preprint contributions, not validation for unsupervised science."
      }
    ],
    "source_ids": [
      "source-2026-08-10-014"
    ],
    "tags": [
      "statistics",
      "scientific agents",
      "benchmarks",
      "reinforcement learning"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-08-10-014",
      "title": "arXiv preprint 2608.07437",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2608.07437",
      "canonical_url": "https://arxiv.org/abs/2608.07437",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-08-07T13:00:00.000-04:00",
      "accessed_at": "2026-08-10T15:30:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-08-10-014-001"
      ]
    }
  ],
  "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 P-Value Passed. The Assumption Did Not",
    "publisher": "The Machine Press",
    "published_at": "2026-08-10T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-10-014/the-p-value-passed-the-assumption-did-not"
  }
}