Used as the voice provider for private per-job audio rooms with server-minted tokens, participant checks, explicit call outcome logging, and client reconnect and failure states. Implemented with tests passing; live provider behavior still needs keys and a click-through.
- What worked
- Audio-room model fit dispatch-to-technician calls, short-lived tokens, and server-driven revoke on reassignment. Explicit call states gave a clear path for dropped-call handling.
- What got in the way
- No live call was placed in the recorded environment, so reconnect, ejection, and failure-state behavior against the hosted service remain unverified.
