Oct 08, 2026 by Offer

Citadel Urn Game 2027: Bayesian Updating Strategy

The Citadel urn betting game — commonly reported by candidates as "An urn has balls of unknown colors. How do you bet optimally as draws are revealed?" — is solved with Bayesian updating: start from a prior over the possible compositions, update the probabilities with each draw, and size your bets to your edge.

What This Citadel Urn Betting Game Question Assesses

Every trading decision is a bet under uncertainty with arriving information, and this game is the stripped-down version. Citadel tests three things: that you start with an explicit prior instead of guessing, that you update it correctly (Bayes' rule, not gut feel), and that your bet sizes reflect your confidence — small edge, small bet.

How to Answer This Citadel Urn Betting Game Question

Play it in three phases:

Phase 1 — Set the prior. With no information, assume all compositions equally likely (uniform prior). State it: "I'll start assuming any mix is equally probable." This is your baseline before data arrives.
Phase 2 — Update on each draw. After each draw, reweight: compositions consistent with what you saw become more likely. Example line: "Three reds in a row — the high-red compositions are now much more probable, so I'll update my estimate of the red fraction upward." If draws are with replacement, use the Beta-Binomial update; narrate the direction even if you approximate the math.
Phase 3 — Bet the edge. Bet only when your estimated probability beats the offered odds, and size proportionally to the edge — a small edge means a small bet. Never bet the same size regardless of confidence; that's the tell of someone who doesn't actually update.

The sentence that scores: "My beliefs are a distribution, not a number — and every draw reshapes it."

Common Mistakes

Ignoring the prior. Starting with a gut guess instead of an explicit prior means your updates have no anchor — and the interviewer will notice.
Over-updating on small samples. Three reds in a row moves the needle but doesn't prove the urn is all red; update proportionally to the evidence.
Flat bet sizing. Betting the same amount at 55% confidence and 90% confidence shows you don't connect belief to action — the core of the game.

Candidates commonly report that the interviewer changes the rules mid-game — draws without replacement, or a cost per draw. The Bayesian framework absorbs all of these: just update the likelihood to match the new rules and keep betting the edge.

Keep Reading

FAQ

What prior should I use? Uniform over compositions is the standard neutral choice. If the interviewer gives base-rate information, fold it in — ignoring given information is a mistake.

How precise does the updating math need to be? Direction and rough magnitude beat exact decimals. "Much more likely now" with correct reasoning outscores a fumbled exact calculation.

Should I use the Kelly criterion for sizing? Mentioning Kelly (bet fraction proportional to edge over odds) shows sophistication, but even "bigger edge, bigger bet" with discipline is enough.

What if draws are without replacement? Then each draw also changes the composition — update using the hypergeometric likelihood instead of binomial. State the adjustment explicitly.

Preparing for Citadel's interview? Our 2027 Citadel Online Assessment and Coding Challenge Tutorials has practice questions and answers — $79 one-time, instant download.