{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-08-21-008",
    "source_story_id": "tmp-story-llm-tabular-anomaly-detector",
    "edition_id": "mp-2026-08-21-morning-0043",
    "edition_url": "https://themachinepress.com/edition/2026-08-21",
    "position": 8,
    "story_type": "dispatch",
    "section": "research",
    "editorial_classification": "editorial",
    "headline": "The Model Wrote the Anomaly Score Instead of Learning It",
    "slug": "the-model-wrote-the-anomaly-score-instead-of-learning-it",
    "dek": "An in-context pipeline turned normal-state summaries into executable scoring logic and beat tested baselines across 24 datasets.",
    "summary": "An in-context pipeline turned normal-state summaries into executable scoring logic and beat tested baselines across 24 datasets.",
    "body_text": "LLM-Detector summarizes normal tabular data into statistics, causal dependencies and prototypes, then asks a language model to generate a scoring engine for deviation, structural inconsistency and density. The authors compare it with 15 baselines across 24 mixed and continuous datasets and report consistent gains without model fine-tuning. The evidence is benchmark-based; generated scoring code still needs review, security controls and domain validation.",
    "why_it_matters": "An in-context pipeline turned normal-state summaries into executable scoring logic and beat tested baselines across 24 datasets.",
    "limitations": [],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-21-008/the-model-wrote-the-anomaly-score-instead-of-learning-it",
    "json_url": "https://themachinepress.com/story/mp-2026-08-21-008.json",
    "first_published_at": "2026-08-21T09:00:00.000-04:00",
    "modified_at": "2026-08-21T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-08-21-008-001",
        "text": "An in-context pipeline turned normal-state summaries into executable scoring logic and beat tested baselines across 24 datasets.",
        "source_ids": [
          "source-2026-08-21-008"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-08-21-008"
    ],
    "tags": [
      "anomaly detection",
      "tabular data",
      "code generation"
    ],
    "image_url": "https://themachinepress.com/issues/2026-08-21/anomaly-code-file-image.webp",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-08-21-008",
      "title": "arXiv preprint 2608.19463",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2608.19463",
      "canonical_url": "https://arxiv.org/abs/2608.19463",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-08-18T20:00:00.000-04:00",
      "accessed_at": "2026-08-21T08:22:06.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-08-21-008-001"
      ]
    }
  ],
  "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 Model Wrote the Anomaly Score Instead of Learning It",
    "publisher": "The Machine Press",
    "published_at": "2026-08-21T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-21-008/the-model-wrote-the-anomaly-score-instead-of-learning-it"
  }
}