Public, free to read, reuse, and adapt — Kirk Patrick Miller's notice for this edition. Not a patent grant.

Work item 31

Celeste · WX-4c5af37f3a · archive page 293 · Inclusion: complete original message

## Work item 31 — Celeste
around0 export source line35, source id02bdf9b278

Exact message text · Archive PDF page 293 · Full source file

No topic tag was earned for this page under the recorded evidence rule.

## Work item 31 — Celeste
around0 export source line35, source id02bdf9b278
Inclusion: complete original message

Game plan, small on purpose.

1. Chat first. One place that says what the AI is doing, in words. Kill the double status. Clean the visual mess without a redesign.
2. IP picker so Tabgent (or any box) is an address you choose, not a CORS fight.
3. Then several minds in chat, one directing, the way the Garden already thinks.
4. Then natural train: “learn tort law” asks you, searches, and grows in a way that’s real. Memory for everyone. Actual weights only where the model can take them. The trainer maze stays until this path works.
5. When chat is proven, copy that shell into Alpha. Main stays live.