{
  "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": "2609.23816",
  "title": "SPLASH: Co-Designing Sparse Attention with High-Bandwidth Flash for Efficient Long-Context Inference",
  "summary": "SPLASH replaces GPU-only HBM KV-cache placement with a virtualized HBM plus High-Bandwidth Flash (HBF) hierarchy, and replaces hardware-agnostic sparse attention with attention co-designed to flash page granularity and plane-level parallelism. The cost is dependence on HBF, an emerging storage substrate not available on standard 24 GB GPUs or API services, plus the engineering complexity of managing a two-tier KV cache and page-aligned sparsity. It can fail through flash write-endurance limits if KV traffic writes excessively, latency spikes from page-granularity mismatches between sparsity patterns and flash reads, and accuracy degradation of up to roughly 4% relative to dense attention.",
  "rationale": "The method requires HBF hardware and systems-level KV-cache control that neither a single 24 GB GPU nor third-party APIs currently expose, so it is relevant to track but not adoptable now.",
  "technique": "HBM-flash KV cache hierarchy with page-aligned sparse attention",
  "family": "cache_kv",
  "recommendation": "observar",
  "published": "2026-09-20",
  "as_of": "2026-10-18",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2609.23816",
  "independent_implementations": 0,
  "total_implementations": 0,
  "stars": 0,
  "citations": 0,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "SPLASH replaces GPU-only HBM KV-cache placement with a virtualized HBM plus High-Bandwidth Flash (HBF) hierarchy, and replaces hardware-agnostic sparse attention with attention co-designed to flash page granularity and plane-level parallelism. The cost is dependence on HBF, an emerging storage substrate not available on standard 24 GB GPUs or API services, plus the engineering complexity of managing a two-tier KV cache and page-aligned sparsity. It can fail through flash write-endurance limits if KV traffic writes excessively, latency spikes from page-granularity mismatches between sparsity patterns and flash reads, and accuracy degradation of up to roughly 4% relative to dense attention.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2609.23816",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "claim_id": "claim-02",
      "statement": "Improves decode throughput per GPU by 3.5x-11.4x over evaluated baselines under a 100 ms per-token latency objective, with accuracy within 4% of dense attention on long-context suites.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2609.23816",
      "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": ""
}