{ "schema_version": 1, "builder_plan": { "scenario": "An agent looks good in demos, but you cannot tell which product tasks it fails.", "change": "Use the study's evaluation idea to inspect a small set of tasks with human-checked answers.", "baseline": "Your current evaluator and a manually reviewed reference set.", "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": "Agreement with human review, missed failures, false alarms, and evaluation cost per task.", "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": "2610.00609", "title": "Legal Research Bench: Measuring End-to-End Reliability in Long-Horizon Legal Research Agents", "summary": "LRB replaces ad hoc or unverified evaluation of legal research agents with 413 expert-written questions graded by binary all-pass rubrics plus citation-source verification, using an LLM judge validated against attorneys. Cost is the benchmark's own setup: a tool harness (web search, case-law search, retrieval) and judge-inference spend per run, plus per-model API usage across thirteen frontier models. What can fail is overfitting tuning decisions to one U.S.-law benchmark, judge drift if the grader model changes, and mistaking benchmark rank for production reliability given that even the best system fails over half of tasks.", "rationale": "The paper contributes an evaluation method rather than a deployable technique, so its value to a constrained team is workload-specific: adopt its all-pass-with-source-verification grading pattern when building legal or other citation-sensitive retrieval agents, but only after validating it against the reader's own task distribution.", "technique": "Legal Research Bench: all-pass agent evaluation benchmark", "family": "avaliacao_de_agente", "recommendation": "testar", "published": "2026-09-30", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.00609", "independent_implementations": 0, "total_implementations": 1, "stars": 10, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "LRB replaces ad hoc or unverified evaluation of legal research agents with 413 expert-written questions graded by binary all-pass rubrics plus citation-source verification, using an LLM judge validated against attorneys. Cost is the benchmark's own setup: a tool harness (web search, case-law search, retrieval) and judge-inference spend per run, plus per-model API usage across thirteen frontier models. What can fail is overfitting tuning decisions to one U.S.-law benchmark, judge drift if the grader model changes, and mistaking benchmark rank for production reliability given that even the best system fails over half of tasks.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.00609", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Best model (Claude Opus 4.8) achieves 42.9% all-pass accuracy; more turns, tool calls, and inference cost do not predict higher accuracy.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.00609", "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": "" }