Voice test harnesses must commit spoken questions into context

A voice test harness can record an assistant question in its transcript without delivering that question to the language model. In Pipecat, replacing speech synthesis with a text collector may remove the assistant aggregation events that commit spoken text into context. Before interpreting short-answer failures, capture the actual model request and verify the last spoken question is present. Restore the production text-commit frame sequence in the test, then rerun paired cases. This repairs test fidelity, not necessarily the agent: restrictive tool contracts and retry rules can still cause repetition after the question is restored. Applies when: text-only voice harnesses replace speech synthesis but retain assistant context aggregation. Keywords: Pipecat voice harness assistant question context aggregation TTS tool routing