{ "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": "2610.29332", "title": "Do Coding Agents Understand Why Systems Crash---and Why They Don't?", "summary": "The method replaces single-version crash prediction benchmarks with paired source trees differing by one commit, so a correct-looking prediction on crashing code alone is no longer accepted as evidence of execution reasoning. It costs the construction of matched pre/post-commit tree pairs with a verified crash-triggering input (934 pairs here from Linux and OSS-Fuzz), plus roughly double the agent evaluation runs per case. It can fail if pairs are not truly minimal, if the crash input is flaky, or if agents refuse or time out on large systems codebases, and it measures crash reasoning only, not general bug-fixing ability.", "rationale": "For a team shipping coding agents on large codebases, adopting paired evaluation is a low-cost way to detect suspicious-pattern shortcutting before trusting agent crash or bug triage in production.", "technique": "Paired source-intervention evaluation for crash reasoning", "family": "avaliacao_de_agente", "recommendation": "testar", "published": "2026-10-21", "as_of": "2026-10-25", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.29332", "independent_implementations": 0, "total_implementations": 1, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces single-version crash prediction benchmarks with paired source trees differing by one commit, so a correct-looking prediction on crashing code alone is no longer accepted as evidence of execution reasoning. It costs the construction of matched pre/post-commit tree pairs with a verified crash-triggering input (934 pairs here from Linux and OSS-Fuzz), plus roughly double the agent evaluation runs per case. It can fail if pairs are not truly minimal, if the crash input is flaky, or if agents refuse or time out on large systems codebases, and it measures crash reasoning only, not general bug-fixing ability.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.29332", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Agents exhibit a bias toward predicting no crash: they rarely raise false alarms but miss many real crashes, with the best agent missing over a quarter of them; agents that merely notice suspicious code predict crashes on both trees, so only paired evaluation reveals this failure.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.29332", "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": "" }