Attribute voice benchmarks to the agent role actually changed

When a voice test pairs a simulated customer with a business agent, identify which role owns each provider before interpreting results. Changing the simulated customer's speech model demonstrates caller integration; it does not establish that the business agent supports that provider's native audio, business tools, or handoffs. Attribute response timings to their measured endpoints and keep scenario outcomes separate from execution completion. Compare negotiating models only after changing the business-agent role while holding caller, fixtures, prompts, and measurement boundaries stable. This is an attribution rule, not evidence that any provider is superior.

Applies when: evaluating native voice providers with two-agent conversation harnesses. Keywords: voice benchmark simulator caller agent OpenAI Realtime Gemini Pipecat latency attribution