Give required voice scripts one speech owner

In a Pipecat flow with continuous model history, both model text and native actions can emit the same required script. If only native emission arms playback tracking, a model-spoken introduction can finish while the flow still marks it owed; identity confirmation then queues it again. Choose one owner for required speech and route recovery through that owner's tracked path. Clear debt only on correlated completion, while ordinary answers retain their own wording. Appending conversation history alone does not coordinate model output with application effects. This addresses speech ownership and framework receipts, not handset audibility or control of every business question.

Applies when: Model speech and flow actions share a voice pipeline with mandatory scripts. Keywords: Pipecat introduction model context speech owner replay completion duplicate