{
  "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": "2609.25678",
  "title": "Toolcompass: Guiding Tool Trialing, Not Suppressing It",
  "summary": "ToolCompass replaces unguided outcome-based post-training of tool-using agents by clustering tool-call representations into von Mises-Fisher distributions per function class, reducing intra-function variation and increasing inter-function separation so exploration transfers to unseen tools. It costs only a modified post-training objective (representation organization losses on tool-call embeddings), requires no ground-truth call traces or unseen-tool access, and adds no inference-time memory or latency. It can fail if the assumed function classes are poorly defined or if the embedding space does not actually reflect functional similarity, in which case the structured representations misdirect exploration on genuinely novel deployment tools.",
  "rationale": "The method is a training-time addition compatible with GRPO/RFT/DMPO that a small team can run on modest models within a 24 GB GPU budget, but its benefit depends on whether the reader's tool distribution matches the benchmark assumptions, so it warrants a workload-specific trial rather than immediate adoption.",
  "technique": "Representation-structured post-training for tool-use agents",
  "family": "uso_de_ferramenta",
  "recommendation": "testar",
  "published": "2026-09-22",
  "as_of": "2026-10-18",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2609.25678",
  "independent_implementations": 0,
  "total_implementations": 0,
  "stars": 0,
  "citations": 0,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "ToolCompass replaces unguided outcome-based post-training of tool-using agents by clustering tool-call representations into von Mises-Fisher distributions per function class, reducing intra-function variation and increasing inter-function separation so exploration transfers to unseen tools. It costs only a modified post-training objective (representation organization losses on tool-call embeddings), requires no ground-truth call traces or unseen-tool access, and adds no inference-time memory or latency. It can fail if the assumed function classes are poorly defined or if the embedding space does not actually reflect functional similarity, in which case the structured representations misdirect exploration on genuinely novel deployment tools.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2609.25678",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "claim_id": "claim-02",
      "statement": "Improves AppWorld out-of-distribution task success by up to 10.71 percentage points over vanilla post-training, with gains reported across GRPO, RFT, and DMPO on AppWorld and FTRL.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2609.25678",
      "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": ""
}