You know the Monty Hall problem. Three doors, one car, two goats, the host reveals a goat, and you should switch because switching doubles your odds from 1/3 to 2/3. It's the most argued-about probability puzzle in history, and most people's intuitions are still wrong about it decades after Marilyn vos Savant's column. Now here's the twist, originally surfaced on Tumblr by user @sigmaleph and brought to the Guardian's puzzle column by Henk Tijms: what if you secretly want one specific goat — a golden goat worth more than the car — and the host doesn't know this? The host opens a door revealing the ordinary goat (not the golden one). Should you switch? The answer is no. You should stick. And the reason this works is a beautiful piece of conditional probability that most people will get wrong for a second time, having barely recovered from getting the original wrong. The key mechanism is information asymmetry: the host's door-opening behavior leaks different information depending on what's behind your door. When you picked the car (1,000 of 3,000 trials), the host reveals the ordinary goat only half the time — 500 cases. When you picked the ordinary goat (1,000 trials), the host never reveals it because he opened the other goat door — 0 cases. When you picked the golden goat (1,000 trials), the host always reveals the ordinary goat — 1,000 cases. So of the 1,500 times the host reveals the ordinary goat, you're sitting on the golden goat in 1,000 of them. Stick: 2/3 win rate. Switch: 1/3. The strategy flips completely. What makes this elegant is that the reversal isn't arbitrary — it falls directly out of how the host's constrained behavior (must avoid the car) interacts with your changed objective. The host is still optimizing around the car. You're not. That mismatch is the entire engine. The puzzle traces to a Tumblr post by @sigmaleph, and the Guardian column credits Henk Tijms for flagging it. It's a small, self-contained piece of recreational mathematics — not a research result, not a longform essay. The column runs on alternate Mondays and has since 2015. This installment earns its place by doing what the best puzzle columns do: making you feel the wrongness of your first instinct twice in the same problem. The frequency approach used in the solution — imagine 3,000 trials and count — is the right pedagogical move. It sidesteps Bayes' theorem notation entirely and lets you see the answer in a table. This is how probability should be taught more often: not as formula manipulation but as counting exercises on large imagined populations. As a cultural artifact, this is a well-executed puzzle column entry. It won't change mathematics. It will change the way a few thousand readers think about conditional probability for the rest of the week, and that's exactly the right ambition for the form.
