{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-06-026",
    "source_story_id": "tmp-story-kv-cache-history",
    "edition_id": "mp-2026-10-06-morning-0089",
    "edition_url": "https://themachinepress.com/edition/2026-10-06",
    "position": 15,
    "story_type": "dispatch",
    "section": "infrastructure",
    "editorial_classification": "editorial",
    "headline": "The Same Prompt Changed After the Cache Saw Different Requests",
    "slug": "the-same-prompt-changed-after-the-cache-saw-different-requests",
    "dek": "Selective KV reuse made rolling-agent answers depend on the order of earlier requests.",
    "summary": "Selective KV reuse made rolling-agent answers depend on the order of earlier requests.",
    "body_text": "Rolling agents repeatedly evict and append documents, breaking exact prefix caching and inviting selective reuse of old key-value states. The study found that persistent reuse could make an unchanged prompt produce different answers depending on the requests processed before it. At a matched 5% recomputation budget, recomputing document-aligned spans reduced order-driven answer variation from 69.0% under token top-k selection to 26.1%. It also improved fidelity to full prefill by 34.5 to 52.5 percentage points while both approaches retained roughly a 5.7-times median time-to-first-token speedup. The cache is not just an optimization when its history can alter the answer.",
    "why_it_matters": "Selective KV reuse made rolling-agent answers depend on the order of earlier requests.",
    "limitations": [
      "The cache is not just an optimization when its history can alter the answer."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-06-026/the-same-prompt-changed-after-the-cache-saw-different-requests",
    "json_url": "https://themachinepress.com/story/mp-2026-10-06-026.json",
    "first_published_at": "2026-10-06T09:00:00.000-04:00",
    "modified_at": "2026-10-06T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-06-026-001",
        "text": "Selective KV reuse made rolling-agent answers depend on the order of earlier requests.",
        "source_ids": [
          "source-2026-10-06-015"
        ],
        "qualification": "The cache is not just an optimization when its history can alter the answer."
      }
    ],
    "source_ids": [
      "source-2026-10-06-015"
    ],
    "tags": [
      "KV cache",
      "rolling agents",
      "inference"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-06-015",
      "title": "arXiv preprint 2610.05833",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2610.05833",
      "canonical_url": "https://arxiv.org/abs/2610.05833",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-05T01:37:22.000-04:00",
      "accessed_at": "2026-10-06T08:25:42.165-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-06-026-001"
      ]
    }
  ],
  "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 Same Prompt Changed After the Cache Saw Different Requests",
    "publisher": "The Machine Press",
    "published_at": "2026-10-06T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-06-026/the-same-prompt-changed-after-the-cache-saw-different-requests"
  }
}