{ "schema_version": 1, "builder_plan": { "scenario": "Your coding assistant makes changes that still need substantial repair before merging.", "change": "Try the technique on held-out issues from a repository with reliable tests.", "baseline": "The same coding agent, repository snapshot, tools, and token budget.", "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": "Correct patches, regression failures, human repair time, and total cost per accepted change.", "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": "2610.30499", "title": "Can LLMs Improve Their Own Harness? A Controlled Study on ARC-AGI-3", "summary": "The method replaces human engineering of the symbolic harness (perception, memory, tool use, recovery, inference allocation) with frontier-model-generated code that the model revises twice, evaluated on held-out ARC-AGI-3 games. It costs full frontier-model inference for harness construction plus two revision cycles across 27 lineages, with dollar-denominated inference cost tracked as a resource measure; no GPU training is required, but the API spend is nontrivial. Self-revisions can fail to generalize (held-out gains are near zero and within noise), can regress existing capabilities, and human-steered harnesses retain substantial headroom.", "rationale": "The study shows autonomous harness self-improvement yields marginal, statistically uncertain held-out gains and risks capability regression, so a constrained team should keep engineering harnesses manually while monitoring this line of research rather than adopting it in production.", "technique": "LLM self-improvement of agent harness code", "family": "agentes_de_codigo", "recommendation": "observar", "published": "2026-10-21", "as_of": "2026-10-24", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.30499", "independent_implementations": 2, "total_implementations": 3, "stars": 6, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces human engineering of the symbolic harness (perception, memory, tool use, recovery, inference allocation) with frontier-model-generated code that the model revises twice, evaluated on held-out ARC-AGI-3 games. It costs full frontier-model inference for harness construction plus two revision cycles across 27 lineages, with dollar-denominated inference cost tracked as a resource measure; no GPU training is required, but the API spend is nontrivial. Self-revisions can fail to generalize (held-out gains are near zero and within noise), can regress existing capabilities, and human-steered harnesses retain substantial headroom.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.30499", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Two self-revision rounds raise mean held-out RHAE by only 2.80 points (95% CI [-1.44, 7.35]); development gains overstate held-out improvement, and cost reductions can degrade task performance.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.30499", "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": "" }