The six sessions 🎯 · one decision memo, built across all of them
Bring a live decision from your own week in session 1. Each session fills one memo section; session 6 delivers the finished memo and freezes the metric that will judge it.
Decisions first
Commit to the default up front, criteria before data, and the six-word question that separates decision-makers from decoration - the Kozyrkov discipline, quoted at source.
🎚️ Session 2 · mediumThe calibration lab
Ten questions, verified answers, your own 90 percent intervals scored live. The number stings, the skill trains, and the memo starts carrying honest estimates.
🦊 Session 3 · mediumSuperforecasting habits
Base rates before case details, small frequent updates, the chimp line taught with its author's own correction - then the forecast game, where a base-rate bot is humblingly hard to beat.
📏 Session 4 · hands-onUncertainty you can promise
From vibes to guarantees: calibration curves, and conformal prediction's distribution-free coverage - with the marginal-not-conditional fine print stated plainly.
💰 Session 5 · hands-onValue of information
Howard's EVPI live in a calculator: the $5k study worth at most $3.6k, the measurement inversion, and how to cancel an analysis with arithmetic instead of politics.
📨 Session 6 · capstoneDecide and communicate
The finished memo: uncertainty stated as ranges (nearly trust-neutral, per the PNAS evidence), the judging metric Goodhart-proofed four ways, the verdict delivered.
Your passport 🎫
Clear all three quiz questions in a session and it gets stamped. Stamps live in this browser only.
Choose your path 🗺️
Three doors into the same discipline: the decision, not the analysis, is the unit of work.
What is real here 🧾
A course about calibrated claims has to calibrate its own.
| Claim | Status |
|---|---|
| The calibration lab | Measures YOU - real hit-rate arithmetic on your own intervals against ten facts, each verified to a cited source that reveals on scoring. No simulation anywhere in it |
| The forecast game | Twelve synthetic scenarios with seeded outcomes, stated on the widget; every Brier number is real arithmetic - yours and both baseline bots (coin-flip 0.250, base-rate 0.171) |
| The value-of-information calculator | Pure live math - change any number and everything recomputes; the do-not-buy verdict at the preset is arithmetic, not opinion |
| The dart-throwing chimpanzee | Quoted WITH Tetlock's own correction: average experts, long-range questions, "'Many' does not equal all" - the garbled version is named and retired |
| "Superforecasters beat analysts with classified data by 30%" | Deliberately not taught as fact - press-sourced, unpublished, aggregation-confounded; the honest version is on the session 3 page |
| The training result | Stated precisely: a ~45-minute module (not "one hour"), with benefits lasting 8-10 month periods, per the tournament paper read in full text |
The knowledge map 🧠
The whole course at a glance - hover a session to spotlight its concepts, click any node to jump in.