{ "schema_version": 1, "builder_plan": { "scenario": "You already need task-specific training and want to reduce its memory or compute cost.", "change": "Change only the training technique on a small controlled run after checking prerequisites.", "baseline": "Your current training recipe with the same model, data split, and evaluation.", "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": "Held-out quality, peak memory, training time, and full compute cost to the target quality.", "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": "2610.00890", "title": "Cross-Benchmark Transfer from RL on Agentic Coding Tasks", "summary": "The method replaces prompt engineering and supervised fine-tuning of coding agents with RL on 1,700 expert-built tasks graded by hidden fail-to-pass and pass-to-pass tests, where reward zeroes out if any existing-behavior test fails. It requires post-training a 1T-parameter (32B active) MoE model with reinforcement learning infrastructure, 1,700 verified tasks, and hidden test harnesses, which exceeds a single 24 GB GPU and a limited cloud budget even though the adapter itself is rank-32 LoRA. The approach can fail through reward hacking against the grader, overfitting to the training harnesses despite reported cross-harness transfer, and regression on existing behavior if the pass-to-pass gating is incomplete.", "rationale": "The evidence is strong that RL post-training improves coding agents, but reproducing it demands foundation-model-scale training infrastructure and curated verified task sets that the reader's constrained setup cannot support, so the practical action is to track whether open-weight checkpoints trained this way become available for direct adoption.", "technique": "RL post-training with LoRA on agentic coding tasks", "family": "treino_eficiente", "recommendation": "observar", "published": "2026-10-01", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.00890", "independent_implementations": 0, "total_implementations": 1, "stars": 10, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces prompt engineering and supervised fine-tuning of coding agents with RL on 1,700 expert-built tasks graded by hidden fail-to-pass and pass-to-pass tests, where reward zeroes out if any existing-behavior test fails. It requires post-training a 1T-parameter (32B active) MoE model with reinforcement learning infrastructure, 1,700 verified tasks, and hidden test harnesses, which exceeds a single 24 GB GPU and a limited cloud budget even though the adapter itself is rank-32 LoRA. The approach can fail through reward hacking against the grader, overfitting to the training harnesses despite reported cross-harness transfer, and regression on existing behavior if the pass-to-pass gating is incomplete.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.00890", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "One epoch of GSPO on a rank-32 LoRA adapter improves pass@1 on all six external benchmarks, e.g. DeepSWE 31.0 to 43.4, Terminal-Bench 2.1 67.4 to 82.0, SWE-Marathon 5.0 to 25.0, with 24-35% shorter median trajectories; gains are absolute point improvements, not multiplicative factors.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.00890", "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": "" }