{ "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": "2608.19072", "title": "What is Missing from AI Post-Training AI: An Empirical Analysis", "summary": "The paper is a diagnostic analysis rather than a deployable method: it replaces the assumption that agent-run post-training pipelines improve strategy over time with evidence that agents lock into an initial strategy and only perform local adjustments, even with added experience scaffolds, human guidance, or inference compute. It costs nothing to adopt directly, but any team automating post-training with LLM agents must budget for human strategy oversight, since no tested intervention restored strategy-level revision. It can fail by being over-read as a solved problem: the scaffold gains are execution-level only, and the hardest tasks show almost no benefit from extra reasoning compute.", "rationale": "The work is a negative-result diagnosis of autonomous post-training agents, so the actionable takeaway is to keep human strategy review in any agent-driven training loop rather than to adopt a new technique.", "technique": "Strategy-level reevaluation in agent-driven post-training", "family": "avaliacao_de_agente", "recommendation": "observar", "published": "2026-08-19", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2608.19072", "independent_implementations": 0, "total_implementations": 1, "stars": 2, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "The paper is a diagnostic analysis rather than a deployable method: it replaces the assumption that agent-run post-training pipelines improve strategy over time with evidence that agents lock into an initial strategy and only perform local adjustments, even with added experience scaffolds, human guidance, or inference compute. It costs nothing to adopt directly, but any team automating post-training with LLM agents must budget for human strategy oversight, since no tested intervention restored strategy-level revision. It can fail by being over-read as a solved problem: the scaffold gains are execution-level only, and the hardest tasks show almost no benefit from extra reasoning compute.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2608.19072", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "An experience-driven scaffold improves execution by +12.6 points on GSM8K and +40.8 on HumanEval, but leaves the agent's training strategy static; the paper quantifies no deployable end-to-end gain from fixing strategy lock-in.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2608.19072", "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": [ { "anchor": "A1.E1", "label": "(1)", "section": "Appendix A Trajectory Analysis Details", "role": "metric", "latex": "C_{G}=\\max_{k\\in\\mathcal{K}}\\frac{1}{|G|}\\sum_{i\\in G}\\mathbf{1}[k_{i,1}=k],", "mathml": "CG=maxk∈𝒦1|G|∑i∈G𝟏[ki,1=k],", "context": "For a group of agent trajectories G with at least one identifiable training strategy, we measure the convergence of initial strategies by", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" }, { "anchor": "A1.E2", "label": "(2)", "section": "Appendix A Trajectory Analysis Details", "role": "metric", "latex": "R_{\\mathrm{change}}(G)=\\frac{\\sum_{(i,t)\\in\\mathcal{P}_{G}}\\mathbf{1}[s_{i,t}\\neq s_{i,t-1}]}{|\\mathcal{P}_{G}|},\\qquad R_{\\mathrm{persist}}(G)=\\frac{\\sum_{(i,t)\\in\\mathcal{P}_{G}}\\mathbf{1}[s_{i,t}=s_{i,t-1}]}{|\\mathcal{P}_{G}|},", "mathml": "Rchange​(G)=∑(i,t)∈𝒫G𝟏[si,t≠si,t−1]|𝒫G|,Rpersist​(G)=∑(i,t)∈𝒫G𝟏[si,t=si,t−1]|𝒫G|,", "context": "Let 𝒫G denote the set of adjacent pairs of training experiments whose strategy states are recognized. We measure the rates at which consecutive experiments persist or change the strategy by", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" } ], "equations_status": "selected", "core_kind": "formula", "equations_fetched_at": "2026-09-18", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }