A Discord co-pilot trained on Jalen's 8-year journal. Grades your setup against the 5-layer framework. Cites the journal entry it learned the pattern from. Refuses signal-calls. Routes broker-truth-backed evidence through the Proof Room.
Setup grading is the surface. Behind it: spiral blocks (refuses bad-pattern setups before grading), override capture (logs your firings + computes the weekly cost), methodology mentor (Q&A grounded in the journal), and a daily framework note that auto-posts pre-open. All four live in the same Discord channel.
Setup matches a logged spiral → grade is refused, prior dates cited.
Fire anyway, override logs. Weekly recap totals cost by skipped layer.
Journal-grounded answers in operator voice. Refuses signal-calls.
Reads today's #daily-framework-note + cites recent matching trades.
The bot doesn't have an opinion on your bias. It has an opinion on whether your bias is defined. Same for the other four — each layer is a question your setup either answers or doesn't.
Direction with reason — HTF, news, prior session. "I think so" doesn't pass.
A named level. VAH, swing high, gap fill. "Around 21420" fails.
The exact event that makes you fire. Not a feeling.
Stop points + position size + percent of account. All three named.
Time-stop. Invalidation. Where you cut before the price-stop.
The co-pilot is a retrieval layer over the journal, with the 5-layer framework as the grader. It cites the journal entry or it stays silent. No LLM wrapper. No prompt over GPT. No scraped forum data.
Every grade traces to a real trade in Jalen's broker statements. You can read every entry the model was trained on — /proof-room exposes statements, /methodology exposes the framework. The model has no separate "secret sauce." The sauce is the journal.
Refresh cadence: trades, overrides, framework refinements feed back nightly. Knowledge cutoff is yesterday's #daily-framework-note. When retrieval returns nothing, the bot says so and tags @operator.