{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-08-28-008",
    "source_story_id": "tmp-story-harnesslens",
    "edition_id": "mp-2026-08-28-morning-0050",
    "edition_url": "https://themachinepress.com/edition/2026-08-28",
    "position": 8,
    "story_type": "dispatch",
    "section": "developer-tools",
    "editorial_classification": "editorial",
    "headline": "The Harness Verified Only the Behaviors It Touched",
    "slug": "the-harness-verified-only-the-behaviors-it-touched",
    "dek": "HarnessLens directs scarce evaluation rollouts toward tasks attributable to each proposed runtime change.",
    "summary": "HarnessLens directs scarce evaluation rollouts toward tasks attributable to each proposed runtime change.",
    "body_text": "Across three agent harnesses and four benchmarks, the authors report held-out gains of 7.6 to 13.6 percent while using less evaluation budget than comparison methods. The framework also gates changes on attributable evidence so an aggregate score cannot as easily bury a targeted regression; the evidence is benchmark-based rather than a production reliability guarantee.",
    "why_it_matters": "HarnessLens directs scarce evaluation rollouts toward tasks attributable to each proposed runtime change.",
    "limitations": [
      "The framework also gates changes on attributable evidence so an aggregate score cannot as easily bury a targeted regression; the evidence is benchmark-based rather than a production reliability guarantee."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-28-008/the-harness-verified-only-the-behaviors-it-touched",
    "json_url": "https://themachinepress.com/story/mp-2026-08-28-008.json",
    "first_published_at": "2026-08-28T09:00:00.000-04:00",
    "modified_at": "2026-08-28T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-08-28-008-001",
        "text": "HarnessLens directs scarce evaluation rollouts toward tasks attributable to each proposed runtime change.",
        "source_ids": [
          "source-2026-08-28-008"
        ],
        "qualification": "The framework also gates changes on attributable evidence so an aggregate score cannot as easily bury a targeted regression; the evidence is benchmark-based rather than a production reliability guarantee."
      }
    ],
    "source_ids": [
      "source-2026-08-28-008"
    ],
    "tags": [
      "agent harnesses",
      "verification",
      "evaluation budgets"
    ],
    "image_url": "https://themachinepress.com/issues/2026-08-28/harnesslens-code-file.webp",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-08-28-008",
      "title": "arXiv preprint 2608.27311",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2608.27311",
      "canonical_url": "https://arxiv.org/abs/2608.27311",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-08-27T12:12:23.000-04:00",
      "accessed_at": "2026-08-28T08:15:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-08-28-008-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "The Harness Verified Only the Behaviors It Touched",
    "publisher": "The Machine Press",
    "published_at": "2026-08-28T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-28-008/the-harness-verified-only-the-behaviors-it-touched"
  }
}