{
  "schema_version": 1,
  "builder_plan": {
    "scenario": "Long customer conversations exhaust memory or limit concurrent requests.",
    "change": "Change only the attention cache policy, after checking model and runtime support.",
    "baseline": "The same model with its ordinary attention cache.",
    "steps": [
      "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.",
      "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.",
      "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.",
      "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test."
    ],
    "measure": "Peak cache memory, long-context answer quality, first-token latency, and requests completed.",
    "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation."
  },
  "builder_review": null,
  "arxiv_id": "2610.21126",
  "title": "Compression Does Not End at Selection: Post-Selection Positional Geometry for Sparse KV Caches",
  "summary": "RoPERelay replaces the default practice of retaining original full-history RoPE coordinates in a sparse KV cache; after selection, it compacts coordinates in an order-preserving way by inverse-rotating and re-rotating cached key phases and synchronizing the current query, without renewed selection, re-prefill, or extra cache budget. It costs roughly 1%-3% additional request latency in the evaluated configurations plus implementation complexity in the inference path, with no extra memory or cache footprint. It can fail if the serving stack does not expose per-key positional manipulation, if the model uses non-RoPE positional schemes, or if gains do not transfer beyond the evaluated long-term memory tasks and selectors to the reader's workloads.",
  "rationale": "The method is inexpensive and compatible with existing selectors and caches, but its quality gains are demonstrated on a narrow set of long-memory tasks and models, so workload-specific validation is warranted before production adoption.",
  "technique": "RoPERelay post-selection RoPE coordinate compaction",
  "family": "cache_kv",
  "recommendation": "testar",
  "published": "2026-10-15",
  "as_of": "2026-10-18",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2610.21126",
  "independent_implementations": 0,
  "total_implementations": 1,
  "stars": 122,
  "citations": 0,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "RoPERelay replaces the default practice of retaining original full-history RoPE coordinates in a sparse KV cache; after selection, it compacts coordinates in an order-preserving way by inverse-rotating and re-rotating cached key phases and synchronizing the current query, without renewed selection, re-prefill, or extra cache budget. It costs roughly 1%-3% additional request latency in the evaluated configurations plus implementation complexity in the inference path, with no extra memory or cache footprint. It can fail if the serving stack does not expose per-key positional manipulation, if the model uses non-RoPE positional schemes, or if gains do not transfer beyond the evaluated long-term memory tasks and selectors to the reader's workloads.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.21126",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "claim_id": "claim-02",
      "statement": "Improves answer quality in 32 of 36 comparisons across two models, six sparse selectors, and three long-term memory tasks; e.g., Llama-3.1-8B-Instruct with BM25 on LongMemEval-S rises from 23.6% to 35.4% accuracy, with approximately 1%-3% complete-request time overhead.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.21126",
      "source_page": null,
      "source_excerpt": ""
    }
  ],
  "exposure_map": [
    {
      "dimension": "quality",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "compute",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "latency",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "operations",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "compatibility",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "security",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "data_and_training",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "reproducibility",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    }
  ],
  "risks": [],
  "minimum_test": [],
  "open_questions": [],
  "independent_tests": [],
  "lab_reproductions": [],
  "report_available": false,
  "validation_tier": "unknown",
  "evidence_tier": "unknown",
  "infrastructure_basis": "unknown",
  "equations": [],
  "equations_status": "not_fetched",
  "core_kind": "",
  "equations_fetched_at": "",
  "equations_source_url": "",
  "equations_source_sha256": "",
  "exposures_reviewed_at": "",
  "exposures_source_sha256": ""
}