{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-08-09-013",
    "source_story_id": "tmp-story-graphrag-overcitation",
    "edition_id": "mp-2026-08-09-morning-0031",
    "edition_url": "https://themachinepress.com/edition/2026-08-09",
    "position": 13,
    "story_type": "dispatch",
    "section": "benchmarks-evals",
    "editorial_classification": "editorial",
    "headline": "The Citation Graph Remembered Too Much",
    "slug": "the-citation-graph-remembered-too-much",
    "dek": "GraphRAG over-cited across corpora, while the damage to answer faithfulness depended on what those extra passages contained.",
    "summary": "GraphRAG over-cited across corpora, while the damage to answer faithfulness depended on what those extra passages contained.",
    "body_text": "Across 4,440 main experiments, 600 cross-corpus runs and 1,200 paired judgments, GraphRAG returned 11 to 15 source IDs per answer with low citation precision in each tested configuration. Faithfulness fell across hops in a typed aviation-requirements corpus but rose on Wikipedia chains because extra passages were still topically useful; even the same judge changed verdicts often when retrieval state changed. The authors argue that architecture claims need tests across embedders, corpora and judges.",
    "why_it_matters": "GraphRAG over-cited across corpora, while the damage to answer faithfulness depended on what those extra passages contained.",
    "limitations": [],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-09-013/the-citation-graph-remembered-too-much",
    "json_url": "https://themachinepress.com/story/mp-2026-08-09-013.json",
    "first_published_at": "2026-08-09T09:00:00.000-04:00",
    "modified_at": "2026-08-09T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-08-09-013-001",
        "text": "GraphRAG over-cited across corpora, while the damage to answer faithfulness depended on what those extra passages contained.",
        "source_ids": [
          "source-2026-08-09-013"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-08-09-013"
    ],
    "tags": [
      "RAG",
      "citations",
      "evaluation"
    ],
    "image_url": "https://themachinepress.com/issues/2026-08-09/graphrag-brain-file-image.webp",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-08-09-013",
      "title": "arXiv preprint 2608.05153",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2608.05153",
      "canonical_url": "https://arxiv.org/abs/2608.05153",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": null,
      "accessed_at": "2026-08-09T08:17:45.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-08-09-013-001"
      ]
    }
  ],
  "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 Citation Graph Remembered Too Much",
    "publisher": "The Machine Press",
    "published_at": "2026-08-09T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-09-013/the-citation-graph-remembered-too-much"
  }
}