Incremental rendering

Readio projects are built as explicit stages:

source/document -> UtterancePlan -> unit WAV cache -> master/timeline -> output

The status command explains freshness from content identities rather than filesystem timestamps:

readio status manuscript.readio
readio status manuscript.readio --json

A changed sentence changes its Utterplan unit content hash, so the next build reuses unchanged unit cache entries. Inserting text can shift positional unit IDs without defeating reuse because the cache key is the unit content hash plus acoustic profile identity. A complete synthesis cache hit does not open a TTS session.

readio render manuscript.readio --format mp3

The high-level command rebuilds only stale stages. Typical decisions are:

readio export tracks each output independently. Its identity includes the master, format, and normalized effective encoder options: M4A defaults to 192k and Opus defaults to 96k. WAV and FLAC do not accept bitrate options. .ogg remains Ogg/Vorbis; .opus is separate.

Audiobook M4B is a separate output identity created by readio audiobook export PROJECT --format m4b. It includes the composition master and timeline, resolved title/author, explicit cover hash, and AAC bitrate (192k by default). Changing only metadata, cover, or bitrate requires a new M4B encode, not replanning, resynthesis, or recomposition. A master/timeline change follows the normal composition invalidation path. Automatic cover extraction is not supported; pass a JPEG/PNG with --cover.

readio preview PROJECT --select paragraph:1-3 --voice VOICE -o preview.wav uses the same synthesis and composition primitives for a selected range. It keeps audition profiles out of the active project trace unless --activate is provided.

Composition writes a standard Audiocompose bundle under composition/ and validates source hashes. Synthesis writes each cache WAV atomically and updates the trace only after all requested audio has been persisted, making interrupted runs resumable.