# Verify audible guidance separately from calibration success

A physical Watch calibration appeared to fail at the left stage, but its persisted timeline proved all captures and the final verified fit succeeded. Speech callbacks advanced stages while James could not hear later instructions. Console playback errors supported an audio problem, without identifying a unique cause.

Keep calibration state, speech lifecycle and human audibility as separate evidence. Log request/start/finish, preserve attempts as distinct files, and label app-session route/volume as such when the synthesizer manages a separate session. Never treat didFinish as proof of audible output. Provide a short audio check before mounting and a prominent current-stage label.

On Watch, build 8 tests system-managed speech using [Apple's usesApplicationAudioSession API](https://developer.apple.com/documentation/avfaudio/avspeechsynthesizer/usesapplicationaudiosession), avoiding repeated media-session activation/deactivation. Physical confirmation remains required; the API is not a promise of a particular output route.

An audio check must not interrupt a pending final calibration announcement: ready state can precede completion of the announcement that starts automatic recording. Guard on pending speech as well as calibration state. A held-completion regression reproduced this failure before the guard and passed afterward.

WatchConnectivity transport completion also differs from application receipt. One archive remained in the phone's system inbox until RideForm was opened; afterward all archive entries matched the source master. Preserve originals and verify receipt before claiming delivery.