{
  "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": "2609.17109",
  "title": "Shared-Prefix KV Reuse Across Standard LoRA Adapters: Quality and Serving Tradeoffs",
  "summary": "The technique replaces per-specialist re-prefilling of a shared context by computing the backbone's prefix KV cache once and reusing it across already-trained standard LoRA adapters. The cost is a small, inconsistent quality degradation on held-out tasks (e.g., -4.6 to -0.8 EM on GSM8K depending on token budget and seed), and the tested implementation copied rather than physically shared KV storage, yielding only 12% lower two-branch peak memory. It can fail when adapter quality degrades beyond tolerance, since no boundary-selection rule or quality equivalence is established and partial recomputation showed no demonstrated benefit.",
  "rationale": "The approach is deployable on constrained single-GPU infrastructure and offers real TTFT savings at long contexts, but the inconsistent quality deltas and absent memory savings mean it must be validated on the reader's specific adapters and workload before production use.",
  "technique": "Shared-prefix KV reuse across LoRA adapters",
  "family": "cache_kv",
  "recommendation": "testar",
  "published": "2026-09-15",
  "as_of": "2026-10-18",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2609.17109",
  "independent_implementations": 0,
  "total_implementations": 0,
  "stars": 0,
  "citations": 0,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "The technique replaces per-specialist re-prefilling of a shared context by computing the backbone's prefix KV cache once and reusing it across already-trained standard LoRA adapters. The cost is a small, inconsistent quality degradation on held-out tasks (e.g., -4.6 to -0.8 EM on GSM8K depending on token budget and seed), and the tested implementation copied rather than physically shared KV storage, yielding only 12% lower two-branch peak memory. It can fail when adapter quality degrades beyond tolerance, since no boundary-selection rule or quality equivalence is established and partial recomputation showed no demonstrated benefit.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2609.17109",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "claim_id": "claim-02",
      "statement": "Warm-cache time-to-first-token improves by roughly 16x at 8K-token contexts when the shared prefix KV is computed once and reused across LoRA specialists, on a Qwen3-1.7B backbone.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2609.17109",
      "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": "no_equations",
  "core_kind": "",
  "equations_fetched_at": "2026-10-18",
  "equations_source_url": "",
  "equations_source_sha256": "",
  "exposures_reviewed_at": "",
  "exposures_source_sha256": ""
}