{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-23-008",
    "source_story_id": "tmp-story-agent-debug-slovak-corpus",
    "edition_id": "mp-2026-10-23-morning-0105",
    "edition_url": "https://themachinepress.com/edition/2026-10-23",
    "position": 8,
    "story_type": "dispatch",
    "section": "open-source",
    "editorial_classification": "editorial",
    "headline": "One Agent Run Becomes a Corpus Test Suite",
    "slug": "one-agent-run-becomes-a-corpus-test-suite",
    "dek": "A read-only auditing agent found seven Slovak web-corpus issues, then compiled them into native-speaker-checked executable tests.",
    "summary": "A read-only auditing agent found seven Slovak web-corpus issues, then compiled them into native-speaker-checked executable tests.",
    "body_text": "Instead of classifying every document with a language model, researchers used a tool-using agent once as a corpus debugger. Three runs over the 26.5-million-document Slovak slice of FineWeb2 cost $2.21 to $2.84 each and produced seven findings across four phenomena. Native-speaker review supported six. The largest was a machine-generated content farm containing 1,408,437 documents, or 5.32 percent of the corpus. The compiled tests then scanned the full slice in 0.7 seconds without a GPU. A static sampling baseline validated more selectors at lower cost, so the agent's value was in prevalence estimates, explanations and reusable tests rather than unqualified superiority.",
    "why_it_matters": "A read-only auditing agent found seven Slovak web-corpus issues, then compiled them into native-speaker-checked executable tests.",
    "limitations": [],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-23-008/one-agent-run-becomes-a-corpus-test-suite",
    "json_url": "https://themachinepress.com/story/mp-2026-10-23-008.json",
    "first_published_at": "2026-10-23T09:00:00.000-04:00",
    "modified_at": "2026-10-23T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-23-008-001",
        "text": "A read-only auditing agent found seven Slovak web-corpus issues, then compiled them into native-speaker-checked executable tests.",
        "source_ids": [
          "source-2026-10-23-008"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-10-23-008"
    ],
    "tags": [
      "web data",
      "corpus quality",
      "agents"
    ],
    "image_url": "https://themachinepress.com/issues/2026-10-23/corpus-code.webp",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-23-008",
      "title": "arXiv preprint 2610.31636, original v1",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2610.31636v1",
      "canonical_url": "https://arxiv.org/abs/2610.31636v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-22T13:52:22.000-04:00",
      "accessed_at": "2026-10-23T08:21:38.024-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-23-008-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "One Agent Run Becomes a Corpus Test Suite",
    "publisher": "The Machine Press",
    "published_at": "2026-10-23T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-23-008/one-agent-run-becomes-a-corpus-test-suite"
  }
}