{
  "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": "2610.20474",
  "title": "AdaRSI: On-Demand Recursive Self-Improvement for Embodied Agents",
  "summary": "AdaRSI replaces indiscriminate fine-tuning of every failing sub-step (or harness-only refinement of a frozen executor) with a failure-attribution loop: a coding agent retries failing sub-steps from saved states, distinguishes harness errors from capability gaps, and fine-tunes only the sub-steps that remain unreliable using self-collected demonstrations. The cost is substantial supporting infrastructure: a checkpointable simulator environment, a coding agent to manage the loop, state saving and replay, and repeated fine-tuning runs of a VLA executor, which is feasible on a single GPU only for modest models. Failure modes include misattribution of harness versus capability errors, low absolute success (13.8% on unseen composites) that limits production readiness, distribution shift from self-collected demonstrations, and dependence on a resettable simulation that most real-world deployments lack.",
  "rationale": "The method is specific to embodied VLA agents in simulation and requires executor fine-tuning plus a full retry-and-attribute harness, so it is relevant to the reader only if they operate robot-learning workloads; otherwise it is a useful design pattern (selective fine-tuning after failure attribution) worth tracking rather than adopting.",
  "technique": "On-demand recursive self-improvement for embodied VLAs",
  "family": "treino_eficiente",
  "recommendation": "observar",
  "published": "2026-10-14",
  "as_of": "2026-10-18",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2610.20474",
  "independent_implementations": 1,
  "total_implementations": 2,
  "stars": 227,
  "citations": null,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "AdaRSI replaces indiscriminate fine-tuning of every failing sub-step (or harness-only refinement of a frozen executor) with a failure-attribution loop: a coding agent retries failing sub-steps from saved states, distinguishes harness errors from capability gaps, and fine-tunes only the sub-steps that remain unreliable using self-collected demonstrations. The cost is substantial supporting infrastructure: a checkpointable simulator environment, a coding agent to manage the loop, state saving and replay, and repeated fine-tuning runs of a VLA executor, which is feasible on a single GPU only for modest models. Failure modes include misattribution of harness versus capability errors, low absolute success (13.8% on unseen composites) that limits production readiness, distribution shift from self-collected demonstrations, and dependence on a resettable simulation that most real-world deployments lack.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.20474",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "claim_id": "claim-02",
      "statement": "On RoboCasa365 with GR00T N1.5 and no human demonstrations, AdaRSI raises success on unseen composite tasks from 4.6% to 13.8% (about 3x), while using 28% of the GPU time and 22% of the demonstrations required by fine-tuning every sub-step.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.20474",
      "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": ""
}