{ "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": "2610.21045", "title": "Learning When to Withdraw: Curriculum Coverage and Rule-Specific Failures in Fine-Tuned Language Agents", "summary": "This is a diagnostic methodology that replaces aggregate accuracy as the acceptance criterion for fine-tuned tool-using agents, adding per-rule probes (wrongful withdrawal with surviving support, narrowing of prohibited combinations, delegation preservation, omitted reports) over held-out dependency families and structural combinations. It costs the construction of a synthetic environment with explicit dependencies and a deterministic reference policy, plus evaluation of each checkpoint under multiple framings; on constrained infrastructure this is feasible since the models used are 3B-parameter LoRA adapters. It can fail by giving false confidence: the authors note that one training run per arm, jointly changed curriculum properties, prior evaluation exposure, and quantization differences prevent causal attribution, and curriculum gaps (e.g., prohibited-combination narrowing) persist even after apparent gains.", "rationale": "The reader can adopt per-rule regression diagnostics on small LoRA fine-tunes immediately, since the paper's central lesson is that aggregate metrics mask specific safety-relevant failures and the required infrastructure is a single-GPU-scale evaluation harness rather than a new model.", "technique": "Rule-level failure diagnostics for fine-tuned agents", "family": "avaliacao_de_agente", "recommendation": "adotar", "published": "2026-10-15", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.21045", "independent_implementations": 1, "total_implementations": 2, "stars": 132, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "This is a diagnostic methodology that replaces aggregate accuracy as the acceptance criterion for fine-tuned tool-using agents, adding per-rule probes (wrongful withdrawal with surviving support, narrowing of prohibited combinations, delegation preservation, omitted reports) over held-out dependency families and structural combinations. It costs the construction of a synthetic environment with explicit dependencies and a deterministic reference policy, plus evaluation of each checkpoint under multiple framings; on constrained infrastructure this is feasible since the models used are 3B-parameter LoRA adapters. It can fail by giving false confidence: the authors note that one training run per arm, jointly changed curriculum properties, prior evaluation exposure, and quantization differences prevent causal attribution, and curriculum gaps (e.g., prohibited-combination narrowing) persist even after apparent gains.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.21045", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Under the naqd prompt, LoRA-SFT checkpoints of Qwen2.5-3B reach decision agreement 0.77-0.92 and withdrawal recall 0.46-0.86 versus 0.45-0.63 agreement and at most 0.06 recall for instruction-tuned baselines; rule-level diagnostics expose regressions (26/34 wrongful withdrawals on the original curriculum, near-zero on the enriched one) that aggregate scores conceal.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.21045", "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": "" }