{ "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": "2609.19334", "title": "A frontend-backend architecture for tool calls in full-duplex speech models", "summary": "The method replaces end-to-end tool-calling inside a full-duplex speech model with a delegation token that routes streaming ASR transcripts to a text LLM backend, injecting results back via a prefill-and-repeat mechanism. The cost is an extra backend LLM in the loop (with its latency and API or serving expense), plus training the frontend to emit delegation tokens and maintaining the transcript-injection plumbing. Failure modes include missed or spurious delegation tokens (roughly 3-8% recall loss and ~19% false-accept on irrelevant requests in their own evaluation), added response latency from the backend round-trip, and degradation of interruption handling if injection races with user barge-in.", "rationale": "Adopting this requires owning and fine-tuning a full-duplex speech frontend, which a constrained-infrastructure team consuming speech via third-party APIs cannot modify, so the architecture is worth tracking rather than deploying now.", "technique": "Frontend-backend delegation for tool calls in duplex speech", "family": "uso_de_ferramenta", "recommendation": "observar", "published": "2026-09-16", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.19334", "independent_implementations": 2, "total_implementations": 3, "stars": 4, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces end-to-end tool-calling inside a full-duplex speech model with a delegation token that routes streaming ASR transcripts to a text LLM backend, injecting results back via a prefill-and-repeat mechanism. The cost is an extra backend LLM in the loop (with its latency and API or serving expense), plus training the frontend to emit delegation tokens and maintaining the transcript-injection plumbing. Failure modes include missed or spurious delegation tokens (roughly 3-8% recall loss and ~19% false-accept on irrelevant requests in their own evaluation), added response latency from the backend round-trip, and degradation of interruption handling if injection races with user barge-in.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.19334", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "92-97% single-turn tool-call recall, 81.2% accuracy at rejecting irrelevant calls, and reported wins over GPT-realtime-mini and Qwen3-Omni-30B-A3B-Instruct on EVA-Bench when paired with a Qwen3-235B-A22B backend.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.19334", "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": "" }