Deploy connectors after #117 (narration voice / beat voice_settings) #119

Activity

1. thorwhalen

   added

   manual-taskRequires the repo owner at the keyboard — agent cannot proceed on its own.Requires the repo owner at the keyboard — agent cannot proceed on its own.

   on Oct 26, 2026
2. thorwhalen commented on Oct 27, 2026

   

   thorwhalen

   on Oct 27, 2026

   CollaboratorAuthor

   More actions

   Four more merges, released as 0.0.78 onward, ride the same connector deploy. Deploy them together, after #117:

   • download_audio: delegate the link-to-audio policy to yb (#122) #126 (download_audio: delegate the policy to yb.download.download_audio_bounded #122): download_audio now delegates to yb.download.download_audio_bounded. The connector venv must resolve yb[download]>=0.1.24. yb is a git or PyPI requirement of braidio[mcp], so confirm that the per-connector refresh actually moves it, not just braidio itself.
   • braidio.Project.genre_status(): the commentary status for a host's one door (#125) #129 (braidio.Project.genre_status(): serve the commentary status through reelee's one project_status #125): braidio.Project.genre_status(). reelee's project_status starts serving the commentary status as soon as the reelee connector's braidio is refreshed. braidio's own project_status tool gains keys (additive).
   • Level spoken parts with mixing.loudness, one measured gain per part (#124) #127 (Level spoken parts with mixing.loudness instead of one-pass loudnorm (two copies) #124): spoken parts are levelled with one measured gain (mixing.loudness). This needs mixing[media]>=0.0.77 in the venv. Rendered audio changes audibly: parts keep their dynamics.
   • A speak= TTS seam: formats render with any voice backend (#123) #131 (A TTS seam so formats (DEEP_DIVE, solo_explainer...) can render with any voxy backend #123): a speak= seam on the Python fast path only. No MCP surface change.

   None of these changes a body schema, so there is no extra=forbid ordering constraint beyond #117's.
3. thorwhalen commented on Oct 27, 2026

   

   thorwhalen

   on Oct 27, 2026

   CollaboratorAuthor

   More actions

   More merges from the same burn ride this deploy:

   • Raise the nw floor to 0.0.108 (tolerant trace readers, nw#184) (#136) #139 (Raise the nw floor to the release carrying nw#184 (tolerant verifying-trace readers) #136): the nw floor is now 0.0.108 (tolerant verifying-trace readers, nw#184). The connector venvs must resolve nw >= 0.0.108. Do this before any project is written by an nw newer than 0.0.107; otherwise an older reader drops the new traces, reads renders as stale, and re-bills them.
   • Graph path takes a voice backend by registered name (#132) #134 (speak= on the graph path (weave_project): a persisted backend id, not a callable #132): dialogue-cast/v1 gains an optional tts field, written only when a voice backend is registered. The connectors register none. Still, deploy the new build before anything writes with it (extra="forbid").
   • A format narrates with its Delivery on the graph too, for new projects (#120) #138 (Fast/graph parity round 4: casting rule, model/settings source, voice spellings (rest of #118) #120): save_script / weave_project with a format now record the format's delivery model (v3) for a project's first weave-config. Projects that already have narration renders keep theirs, so nothing re-bills. The decision is open on Decide: should the graph path use the format's Delivery (re-bills old projects)? #135.
   • Stings: one measured gain + true-peak limit (#130, part 1) #133 (Master and sting levelling: one measured gain instead of one-pass loudnorm #130, part 1): the sting's levelling changed. Durations are unchanged.