{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-17-005",
    "source_story_id": "tmp-story-contiguous-kv-cache-repair",
    "edition_id": "mp-2026-09-17-morning-0070",
    "edition_url": "https://themachinepress.com/edition/2026-09-17",
    "position": 5,
    "story_type": "dispatch",
    "section": "infrastructure",
    "editorial_classification": "editorial",
    "headline": "A Contiguous Window Repaired the Stale Cache",
    "slug": "a-contiguous-window-repaired-the-stale-cache",
    "dek": "Edit-local recomputation recovered at least 0.94 of the post-edit answer margin and ran 13–21× faster than a full prefill.",
    "summary": "Edit-local recomputation recovered at least 0.94 of the post-edit answer margin and ran 13–21× faster than a full prefill.",
    "body_text": "Editing a retrieved document can invalidate downstream key-value states even when the change is local. Across three model families and a factual retrieval benchmark, a contiguous window around the edit recovered at least 0.94 of the post-edit answer margin at the primary budget. Attention, KV-deviation and structural selectors did worse because scattered positions inherited surrounding staleness during real recomputation. Repair ran 13–21 times faster than full re-prefill. The advantage largely disappeared when answer-bearing text moved farther downstream, defining a practical boundary for the method.",
    "why_it_matters": "Edit-local recomputation recovered at least 0.94 of the post-edit answer margin and ran 13–21× faster than a full prefill.",
    "limitations": [],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-17-005/a-contiguous-window-repaired-the-stale-cache",
    "json_url": "https://themachinepress.com/story/mp-2026-09-17-005.json",
    "first_published_at": "2026-09-17T09:00:00.000-04:00",
    "modified_at": "2026-09-17T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-17-005-001",
        "text": "Edit-local recomputation recovered at least 0.94 of the post-edit answer margin and ran 13–21× faster than a full prefill.",
        "source_ids": [
          "source-2026-09-17-005"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-09-17-005"
    ],
    "tags": [
      "KV cache",
      "retrieval",
      "inference efficiency"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-17-005",
      "title": "arXiv preprint 2609.17983",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.17983",
      "canonical_url": "https://arxiv.org/abs/2609.17983",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-09-15T21:13:04.000-04:00",
      "accessed_at": "2026-09-17T08:28:34.205-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-17-005-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "A Contiguous Window Repaired the Stale Cache",
    "publisher": "The Machine Press",
    "published_at": "2026-09-17T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-17-005/a-contiguous-window-repaired-the-stale-cache"
  }
}