{
  "schema_version": 1,
  "builder_plan": {
    "scenario": "An assistant chooses the wrong API or sends arguments your product cannot use.",
    "change": "Apply the proposed tool-selection or argument-generation change in a sandbox.",
    "baseline": "Your current agent with the same tool schemas and task 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 completed actions, invalid calls, retries, latency, and cost per completed 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": "2609.29050",
  "title": "SLCA-GRPO: Resolving Cross-Segment Credit Misattribution in Tool-Calling RL",
  "summary": "SLCA-GRPO replaces standard GRPO's trajectory-level scalar advantage broadcast with segment-decoupled advantage estimation that routes execution rewards to tool-call tokens and preference rewards to summary tokens, trained against a schema-guided LLM simulator instead of real APIs. The cost is substantial infrastructure: building the SGLS simulator, implementing hierarchical rewards and segment-level advantage computation, and running multi-rollout RL training on a 7B model, which strains a single 24 GB GPU and a small team. Failure modes include simulator-to-real API distribution shift that invalidates the learned tool policy, incorrect segment boundaries reintroducing credit misattribution, and gains that may not transfer to smaller backbones or different tool schemas.",
  "rationale": "The method targets teams already performing RL post-training of tool-calling models with dedicated simulator infrastructure, which exceeds the reader's constrained compute and budget, so it is worth tracking rather than adopting or testing now.",
  "technique": "Segment-locked credit assignment for tool-calling RL",
  "family": "uso_de_ferramenta",
  "recommendation": "observar",
  "published": "2026-09-24",
  "as_of": "2026-10-18",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2609.29050",
  "independent_implementations": 0,
  "total_implementations": 1,
  "stars": 3,
  "citations": 0,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "SLCA-GRPO replaces standard GRPO's trajectory-level scalar advantage broadcast with segment-decoupled advantage estimation that routes execution rewards to tool-call tokens and preference rewards to summary tokens, trained against a schema-guided LLM simulator instead of real APIs. The cost is substantial infrastructure: building the SGLS simulator, implementing hierarchical rewards and segment-level advantage computation, and running multi-rollout RL training on a 7B model, which strains a single 24 GB GPU and a small team. Failure modes include simulator-to-real API distribution shift that invalidates the learned tool policy, incorrect segment boundaries reintroducing credit misattribution, and gains that may not transfer to smaller backbones or different tool schemas.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2609.29050",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "claim_id": "claim-02",
      "statement": "On a 7B backbone, SLCA-GRPO outperforms standard GRPO, ToolPO, and RLTR by +2.53 pp in-domain, +1.36 pp on BFCL, and +9.15 pp on tau2-Bench under equal training budgets, with reduced tool redundancy and cost.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2609.29050",
      "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": ""
}