Verify partial text reaches clients before wiring a live transcript

A streaming recognition API can still expose text only after silence or completed inference. Before adding a live transcript UI, trace when audio triggers recognition, whether intermediate text escapes the inference worker, and which callback receives ordinary updates. Listening only for interim frames can miss callback-only updates. Preserve the existing finalized-turn contract and keep provisional display text separate from conversational input. This applies to checkpoint-based recognition behind normalized voice gateways; it does not establish the behavior of an uninspected deployed adapter.

Applies when: A transcript must appear while a speaker is still talking rather than after a pause. Keywords: live transcript interim text recognition checkpoint silence callback streaming