P.4 Audience, Projection, and Caption Records

Audience phones

• Share the Show's mobi.live/p/XXXXXX URL or its QR code.
• Each Viewer selects one available caption language and receives only that output.
• Viewer count does not multiply recognition or translation requests.
• End turns access off between Performances without discarding the Show's URL.
• Renew an expired URL before starting; retrying a collision or cancelled change does not create another Show or Run.

Reading fast dialogue

Portrait uses a bottom-anchored rolling caption. Landscape uses a centered Conversation Stack rather than left/right speaker columns. The latest caption stays white at the bottom, separated by one blank line from a bounded history that becomes lighter as it moves upward.

Speaker diarization may help preserve turn boundaries internally, but uncertain speaker identity never becomes a guessed character name on the Audience screen.

Projection and optional CSV

• Browser projection uses the same accepted caption state and shared projection controls.
• Open the real projector window and send its test caption before starting the Run.
• The optional CSV records committed captions with time relative to Start Listening.
• The CSV does not expose internal segment_id values.
• The CSV excludes raw audio, provisional text, provider payloads, and speaker embeddings.
• If a transcript is incomplete, Summary identifies the uncertain range when possible and confirms whether already shown captions were affected.