{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-11-05-008",
    "source_story_id": "tmp-story-frontierphysics",
    "edition_id": "mp-2026-11-05-morning-0118",
    "edition_url": "https://themachinepress.com/edition/2026-11-05",
    "position": 8,
    "story_type": "dispatch",
    "section": "benchmarks-evals",
    "editorial_classification": "editorial",
    "headline": "Physics Agents Know More Than They Can Prove",
    "slug": "physics-agents-know-more-than-they-can-prove",
    "dek": "FrontierPhysics asks agents to complete 56 end-to-end research tasks authored by 42 physicists, then scores correctness and publishability.",
    "summary": "FrontierPhysics asks agents to complete 56 end-to-end research tasks authored by 42 physicists, then scores correctness and publishability.",
    "body_text": "Each task spans literature review, planning, analysis or experiment, structured output and a scientific manuscript. The benchmark covers nine theoretical, computational and experimental domains and combines deterministic checks with expert-written rubrics. Across 22 models in 26 configurations, no configuration passes more than 46 of 168 rollouts. Manual trajectory review attributes the largest deficits to rigor and expert intuition rather than missing physics knowledge or tool execution. The benchmark measures the supplied research tasks, not autonomous discovery in every scientific field.",
    "why_it_matters": "FrontierPhysics asks agents to complete 56 end-to-end research tasks authored by 42 physicists, then scores correctness and publishability.",
    "limitations": [
      "The benchmark measures the supplied research tasks, not autonomous discovery in every scientific field."
    ],
    "importance": 9,
    "canonical_url": "https://themachinepress.com/story/mp-2026-11-05-008/physics-agents-know-more-than-they-can-prove",
    "json_url": "https://themachinepress.com/story/mp-2026-11-05-008.json",
    "first_published_at": "2026-11-05T09:00:00.000-05:00",
    "modified_at": "2026-11-05T09:00:00.000-05:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-11-05-008-001",
        "text": "FrontierPhysics asks agents to complete 56 end-to-end research tasks authored by 42 physicists, then scores correctness and publishability.",
        "source_ids": [
          "source-2026-11-05-008"
        ],
        "qualification": "The benchmark measures the supplied research tasks, not autonomous discovery in every scientific field."
      }
    ],
    "source_ids": [
      "source-2026-11-05-008"
    ],
    "tags": [
      "physics",
      "research agents",
      "evaluation",
      "scientific rigor"
    ],
    "image_url": "https://themachinepress.com/issues/2026-11-05/frontierphysics-lab.webp",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-11-05-008",
      "title": "arXiv preprint 2611.06910, original v1",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2611.06910v1",
      "canonical_url": "https://arxiv.org/abs/2611.06910v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-11-04T13:56:22.000-05:00",
      "accessed_at": "2026-11-05T08:29:24.413-05:00",
      "supports_claim_ids": [
        "claim-mp-2026-11-05-008-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Physics Agents Know More Than They Can Prove",
    "publisher": "The Machine Press",
    "published_at": "2026-11-05T09:00:00.000-05:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-11-05-008/physics-agents-know-more-than-they-can-prove"
  }
}