{ "schema_version": 1, "builder_plan": { "scenario": "Your inference service slows down when several customers send requests together.", "change": "Replay a recorded request trace with only the proposed serving policy changed.", "baseline": "Your current scheduler on the same model and hardware.", "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": "Requests meeting your latency target, throughput, failures, and cost per successful request.", "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": "2609.31341", "title": "The Right Information Extraction Pipeline Depends on the Document: Accuracy-Energy Trade-offs for Small, Local Models", "summary": "Replaces one-size-fits-all cloud or vision-only extraction pipelines with locally served small models whose input representation (page images vs. parsed text) is chosen per document type, with batching as the primary efficiency lever and classical rather than neural OCR for preprocessing. Costs are the engineering effort of a document classifier or routing rule, batching infrastructure with its latency-throughput trade-off, and maintaining two pipeline variants; quantization adds marginal benefit once batching is in place. Can fail when documents fall between the near-plain-text and layout-rich extremes, when batching is infeasible due to low request volume or strict latency SLOs, or when the cheap parser degrades on slightly more complex layouts, silently dropping accuracy.", "rationale": "The reader has exactly the target setting (single 24 GB GPU, privacy-sensitive or cost-constrained local serving of <=8B models), and the paper's dominant, free win is batching plus routing by document type, both immediately deployable without new hardware or training.", "technique": "Batching-first, document-adaptive local IE pipelines", "family": "serving_e_batching", "recommendation": "adotar", "published": "2026-09-25", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.31341", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "Replaces one-size-fits-all cloud or vision-only extraction pipelines with locally served small models whose input representation (page images vs. parsed text) is chosen per document type, with batching as the primary efficiency lever and classical rather than neural OCR for preprocessing. Costs are the engineering effort of a document classifier or routing rule, batching infrastructure with its latency-throughput trade-off, and maintaining two pipeline variants; quantization adds marginal benefit once batching is in place. Can fail when documents fall between the near-plain-text and layout-rich extremes, when batching is infeasible due to low request volume or strict latency SLOs, or when the cheap parser degrades on slightly more complex layouts, silently dropping accuracy.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.31341", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Batching cuts energy per page by 38-85% at no accuracy cost; FP8 quantization saves 27-32% unbatched but only 9-19% after batching; neural OCR costs 17x more energy per page than classical OCR.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.31341", "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": "" }