{
  "schemaVersion": "2.0",
  "id": "CL-0015",
  "version": 1,
  "recordUrl": "https://cortist.com/reality-ledger/cl-0015/",
  "snapshotUrl": "https://cortist.com/data/reality-ledger/records/cl-0015-v1.json",
  "sourceDate": "2026-11-03",
  "snapshotDate": "2026-11-05",
  "topic": "AI Chips & Compute",
  "state": "documented",
  "status": "Technical-report findings; cross-provider replication pending",
  "organization": "NVIDIA-affiliated research authors",
  "claim": "Can short-context compaction match long-context agent accuracy with lower real-world serving costs?",
  "evidence": "The November 3 arXiv technical report 'You don't need long context' describes a compaction strategy under which selected frontier open-weight models maintained much of their performance across reported tasks. The authors estimate later tokens can be as much as seven times costlier than earlier tokens under their studied serving assumptions. The experiment includes long-horizon software engineering tasks.",
  "limitations": "The sevenfold figure is an upper-bound cost comparison in a particular compute model, not a universally measured API bill reduction. Task selection, model architecture, inference configuration and compaction strategy may affect conclusions. The report is not an independently replicated cost study.",
  "nextEvidence": "Independent reruns reporting matched accuracy, latency, GPU utilization and actual cost per completed task across providers and context policies.",
  "sources": [
    {
      "name": "arXiv: You don't need long context (NVIDIA technical report)",
      "url": "https://arxiv.org/abs/2611.05306",
      "published": "2026-11-03"
    }
  ],
  "timeline": [
    "2026-11-03 | Technical preprint published | Authors report selected short-context task performance and up-to-sevenfold later-token cost effects.",
    "2026-11-05 | Cortist baseline | Independent provider-level cost and accuracy replication outstanding."
  ],
  "newsUrl": "https://cortist.com/news/nvidia-long-context-compaction-sevenfold-token-cost-2026/",
  "topicUrl": "https://cortist.com/ai-chips-and-compute/",
  "kind": "independent_replication",
  "deadline": null,
  "verificationCriteria": [
    "Compare identical model-task settings with/without compaction under reproducible workloads.",
    "Report compute and end-to-end dollars per successful task, not only nominal token cost.",
    "Rerun across multiple providers, open-weight models and long-horizon evaluation sets."
  ],
  "sourceReviewNote": "Cortist read the cited original publication before opening the case; this is not independent replication or resolved-outcome verification",
  "independentlyResolved": false,
  "baselineStatusAsOf": "2026-11-05"
}