{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-08-23-009",
    "source_story_id": "tmp-story-swe-bench-science",
    "edition_id": "mp-2026-08-23-morning-0045",
    "edition_url": "https://themachinepress.com/edition/2026-08-23",
    "position": 9,
    "story_type": "dispatch",
    "section": "benchmarks-evals",
    "editorial_classification": "editorial",
    "headline": "The Best Coding Agent Solved Fewer Than Half",
    "slug": "the-best-coding-agent-solved-fewer-than-half",
    "dek": "SWE-bench Science spans 119 tasks, 98 repositories and twenty fields where software is part of the instrument.",
    "summary": "SWE-bench Science spans 119 tasks, 98 repositories and twenty fields where software is part of the instrument.",
    "body_text": "The benchmark separates issue-driven, expert-exploratory and engineering-integration work and reports a best pass-at-one below 50 percent. Its error analysis finds failures in scientific abstraction, exploration, repair coverage and generalization; a paired ablation also showed that well-grounded scientific guidance can help while poorly aligned guidance can anchor the repair in the wrong direction.",
    "why_it_matters": "SWE-bench Science spans 119 tasks, 98 repositories and twenty fields where software is part of the instrument.",
    "limitations": [],
    "importance": 9,
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-23-009/the-best-coding-agent-solved-fewer-than-half",
    "json_url": "https://themachinepress.com/story/mp-2026-08-23-009.json",
    "first_published_at": "2026-08-23T09:00:00.000-04:00",
    "modified_at": "2026-08-23T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-08-23-009-001",
        "text": "SWE-bench Science spans 119 tasks, 98 repositories and twenty fields where software is part of the instrument.",
        "source_ids": [
          "source-2026-08-23-009"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-08-23-009"
    ],
    "tags": [
      "coding agents",
      "scientific software",
      "benchmarks"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-08-23-009",
      "title": "arXiv preprint 2608.19799",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2608.19799",
      "canonical_url": "https://arxiv.org/abs/2608.19799",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-08-20T04:53:15.000-04:00",
      "accessed_at": "2026-08-23T08:27:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-08-23-009-001"
      ]
    }
  ],
  "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 Best Coding Agent Solved Fewer Than Half",
    "publisher": "The Machine Press",
    "published_at": "2026-08-23T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-23-009/the-best-coding-agent-solved-fewer-than-half"
  }
}