Step 1
Guess
A Fermi question appears: something like “How many dentists work in the United States?” Each player privately submits a range estimate.
Step 2
Bet & Hints
Three betting rounds, interspersed with two progressive hints. The challenge is to update to the new information by the right amount and act on it by risking more or giving up: overreaction is just as costly as failing to move when the evidence genuinely warrants it.
Step 3
Answer & Final Bet
The correct answer is revealed. One last betting round. You now know the truth. Bet accordingly.
Step 4
Showdown
All ranges are revealed. The narrowest range containing the answer wins the pot. If no range contains it, the closest estimate wins.
World-Modeling
Every Fermi question is a question about the structure of the world. How dense are cities? What products do people buy? How do humans actually spend their time? The game rewards players who carry around a rich, roughly-accurate model of reality, and it quietly builds that model in players who don’t yet have one.
Fermi Estimation
The core of the game is decomposition: turning a question you cannot possibly answer directly into a chain of simpler questions you can. This is the central skill of quantitative reasoning, and it is dramatically undertaught. Most people never learn that the question “How many piano tuners are in Chicago?” is not a trivia question at all, but an invitation to think structurally.
Calibration
Your range is a statement about your own uncertainty. Too narrow and you risk missing entirely. Too wide and you often cannot win. Over many rounds, you develop something genuinely valuable: an accurate sense of what you know and what you don’t. Research on expert judgment consistently finds that primarily calibration separates good forecasters from poor ones.
Bayesian Updating
Each hint is a small piece of evidence. The interesting question is not whether it changes your confidence, but how much. Players who learn to update fluidly, adjusting proportionally to the actual evidential weight of new information, have a persistent advantage. Players who anchor to their first number do not.
Cognitive Empathy
A raise after the second hint means something. A fold after a long pause means something different. Players who learn to model other minds find themselves playing a richer game than the one on the surface. And unlike conventional poker, the relevant information extends beyond betting patterns. A question about car insurance costs plays differently against a car enthusiast than against a college student who has never owned a car. Part of reading the table is reading what life experience each player brings to each particular question.
Inherently Social
Built-in video chat keeps the table alive. But the deeper point is that the questions themselves are interesting. They reach into geography, economics, biology, infrastructure, human behavior. Rounds end in genuine arguments about how the world works — the kind of conversations that most games never produce.
Playing in person and don’t need lobby, betting, or video? Launch a streamlined mode where we provide the questions, hints, and answers. You only need some chips, something to write down guesses, and at least one other person physically with you.