Set up poker-coach session-1 baseline
The CLI lives at `~/poker-decision-support/`. Open a terminal, source the venv, and run `python -m poker_coach.db.schema` to confirm the eight tables exist. Then `session start`. The premortem prompts will feel awkward the first time — Annie Duke's discipline is a muscle. The point is to capture three branches and a kill criterion, not to predict the future. Then play the session. Capture one read. Resolve it. Run `calibrate`. The first Brier score is a baseline, not a verdict.
Chapter focus Run your first poker-coach session start to session end, log a hand, log a read, resolve it, and verify the Brier score.
Load-bearing idea
The audit track is the connective tissue of the lab. Without it, you have study (theory, solver) and play (population, bankroll) and discipline (mental), but no measurement. The audit track measures whether the other tracks are working. Brier score on calibrated reads is the primary measurement; EV/100 over rolling 5k hands is the secondary; leak count opened vs closed is the tertiary.
The first session through poker-coach establishes the baseline. Subsequent weekly audits compare against it. The baseline is not flattering. That is the point.
Retrieval prompts
- What are the eight tables in the poker-coach SQLite schema?
- What does
session startcapture beyond stake and format? - What is the Brier-score formula and what range does it span?
- What does Tetlock call “superforecaster range” on a Brier scale?
Re-derivation log
After your first session start-to-end, paste the output of python -m poker_coach.cli calibrate into ~/poker-curriculum/audits/2026-W18-baseline.md. Note the resolved-reads count and the Brier score. Add a one-paragraph reflection on what surprised you, what felt forced, and what you would change about the premortem prompts.