How feedback will work
Feedback is a noun too. People rate what they see; learners use those ratings to guide future work, and never promote anything on ratings alone.
First design, October 29, 2026. Not built yet: today, feedback goes through the feedback form, which needs a free GitHub account.
The noun: one feedback record
Each record holds:
- the target: the exact noun or verb it is about, with its version;
- who gave it;
- a rating, from −1 to +1, or on a 1 to 5 scale;
- optional aspects, such as correct, useful, style or speed;
- a short comment, with a length limit;
- when it was given.
Feedback is private by default. The target’s owner can read it, and so can a learner that holds a grant to.
The verbs: work that uses it
- Request feedback (
request_feedback) - Attaches a request to anything delivered, such as a rendered scene or a finished kit.
- Triage feedback (
triage_feedback) - A worker groups the comments into themes.
- Apply feedback (
apply_feedback) - Creates follow-up verbs, for example “re-render scene 3 with slower captions”.
The surface: where people give it
A thumbs control with a one-line comment on each progress or delivery page, on the home portal or on a progress link.
The learners: what may change
- The preference learner uses ratings to steer future palettes and styles.
- The routing learner uses outcomes plus ratings to choose models and verbs.
- Neither one ever promotes anything on ratings alone.
The infrastructure it needs
- One rating per person for each version of a target.
- The same permission checks as everything else.
- Limits on abuse that are fair to each network, rather than one global cap.