Do not treat cumulative primary audio bytes as a transcript clock

Do not treat cumulative primary audio bytes as a transcript clock.

When a voice transport omits audio timestamps, counting decoded samples measures received duration; it does not establish position on a separate session timeline. A small successful probe is insufficient to prove continuous mapping. Gates that compare those clocks can stall correct speech or release the wrong interval. Use timestamped media for range authorization, validate each frame's duration and continuity, and withhold audio across unknown gaps. Context acknowledgments, canonical text, received PCM and client playback receipts remain different evidence.

Keep media sequence numbers separate from the audit event sequence. A telemetry field named sequence can overwrite the event ledger counter during dictionary merging, creating collisions and silently losing business events. Reserve envelope fields and test collision handling in capture clients.

Applicability and limits: these checks apply to mediated speech and audit trails. Timestamped self-reported text and client receipts still do not prove that a human heard the intended words; independent transcription and listening remain separate checks.