{ "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": "2610.23506", "title": "MLOps-Bench: Benchmarking AI Agents on Production ML Engineering in Real Repositories", "summary": "MLOps-Bench replaces generic issue-fixing or competition-style ML agent evaluations with 212 executable lifecycle tasks (data pipelines, training, serving, monitoring, CI/CD, governance) drawn from 31 real repositories, each with oracle patches, buggy baselines, and fail-to-pass/pass-to-pass tests. Adoption costs only the infrastructure to run repository environments and agent rollouts, fitting a single-GPU or API-based setup, but requires engineering effort to integrate the harness into internal agent selection. It can fail as a decision tool if its public repositories diverge from the reader's private stack, and pass rates may not transfer to proprietary codebases or frameworks outside the benchmark's coverage.", "rationale": "The reader deploying coding or maintenance agents on ML infrastructure can use this benchmark to validate agent capability on production-relevant tasks before adoption, but should confirm task distribution matches their own stack rather than adopting it unconditionally.", "technique": "Repository-grounded benchmark for MLOps agents", "family": "avaliacao_de_agente", "recommendation": "testar", "published": "2026-10-16", "as_of": "2026-10-19", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.23506", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "MLOps-Bench replaces generic issue-fixing or competition-style ML agent evaluations with 212 executable lifecycle tasks (data pipelines, training, serving, monitoring, CI/CD, governance) drawn from 31 real repositories, each with oracle patches, buggy baselines, and fail-to-pass/pass-to-pass tests. Adoption costs only the infrastructure to run repository environments and agent rollouts, fitting a single-GPU or API-based setup, but requires engineering effort to integrate the harness into internal agent selection. It can fail as a decision tool if its public repositories diverge from the reader's private stack, and pass rates may not transfer to proprietary codebases or frameworks outside the benchmark's coverage.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.23506", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Best of six evaluated agents solves 31.2% of 212 production MLOps tasks while passing 68.7% of tests, indicating partial progress without full behavioral correctness.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.23506", "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": "" }