The Missing Review Studio

The Missing Review Studio: What Choir Still Needs for Podcasts and YouTube

Status: published\ Scope: current gap between June source/media substrate and Yusef’s Mosiah review workflow\ Revision reviewed: go-choir main at bff823d7

The most useful short-term Choir product may be very concrete:

paste a YouTube or podcast link
-> pull transcript and metadata
-> help me read/listen with source memory
-> let me write a review in Texture
-> publish it safely to Mosiah or Choir’s public surface

June moved the codebase closer to that, but did not finish the loop.

What already exists

Choir now has pieces that matter for this workflow.

Link and source substrate

The code has a general content/source substrate. YouTube URLs can be detected and normalized. Imported content can carry source type, media type, canonical URL, metadata, transcript availability, and private provenance.

Texture can register pasted media URLs as source entities. That means the writing surface can know that a link is not just a string; it is a source object with possible transcript and media affordances.

YouTube transcript attempts

There is partial YouTube transcript support. The code can try a configured transcript provider, fall back to YouTube player/caption paths, and store transcript text as a derived content item with segment-like metadata.

This is enough to prove the direction: a video link can become a source plus a transcript artifact.

Podcast feed/library substrate

There are podcast types, store methods, runtime handlers, frontend parsing helpers, subscription refresh paths, and media playback state. The system can represent podcast subscriptions and episodes better than a bare URL list.

Texture provenance fields

Texture provenance has fields for source entities, transcript availability, transcript content IDs, timestamp-like selectors, source representation IDs, and publication/export policy. The vocabulary is close to what a review workflow needs.

What is missing

The gap is not philosophical. It is product-loop concrete.

1. No polished review studio

There is no obvious complete path from pasted media link to a ready review Texture with video/audio, transcript, notes, excerpts, citations, and publication controls.

The substrate exists. The surface does not yet feel like “this is where I review an episode.”

2. Podcast transcripts are not first-class

YouTube captions are one path. Podcast audio needs another: audio download or stream reference, speech-to-text, language/speaker/time metadata, retry state, cost policy, and transcript refresh.

The object graph can name audio_recording and transcript, but the product path from podcast episode to transcript-backed review still needs to be built.

3. Transcript span UX is missing

For review writing, the key unit is often a moment:

episode at 37:12–39:04
video claim at 12:30
transcript paragraph with speaker label

The system needs selection, playback sync, excerpt cards, source blocks, and safe citation/export behavior. Metadata fields alone are not enough.

4. Researcher compression is not productized

A good review assistant should not dump an entire transcript into a prompt. It should create compact source representations: episode outline, claims, quotes, counterclaims, timestamps, open questions, and novelty notes.

Choir has the doctrine for this. It still needs the reliable appagent/researcher loop that makes it mundane.

5. Publication safety needs a review-specific proof

A published review should cite and quote sources without leaking private transcripts, raw captures, or paid/private material. The general source/publication policy exists, but a specific YouTube/podcast review acceptance test would make the workflow trustworthy.

The right next target

The next practical Choir milestone should be a narrow review studio v0:

Paste one YouTube URL
-> import video metadata and transcript if available
-> create/open a Texture review document
-> show video + transcript side by side
-> let user select timestamped excerpts
-> insert source-backed quote cards into the review
-> publish draft to Mosiah/Choir with safe source policy

After that, add podcast RSS episodes and speech-to-text.

Why this matters

This is not a side feature. It is the everyday version of the whole Choir thesis.

A review is a small artifact loop: source, attention, notes, claims, revision, evidence, and publication. If Choir can make that loop smooth for one person reviewing what they listen to, then the larger architecture becomes grounded in use rather than doctrine.

The automatic newspaper and the human review studio should share a substrate. Universal Wire reads the world to publish machine-written articles. Mosiah-style review reads the world to help a person write.

Both need Texture. Both need sources. Both need provenance. Both need publication boundaries.