Two doors left. It is not fifty-fifty.

The pull of the wrong answer is almost irresistible. Two doors remain, one hides a car, so surely it is a coin flip — 50/50, switch or stay, who cares. That reasoning has fooled mathematics professors, and it is wrong.

The key is that your door was chosen at a moment when it had a one-in-three chance of being right, and nothing the host does afterwards can reach back and change that. Two times in three, your first pick is a goat. The host then clears away one of the remaining goats — and because your door stays stuck at one-in-three, the entire remaining two-in-three collapses onto the single door you didn't pick and he didn't open. Switching doesn't give you one of two equal options. It gives you the combined odds of both doors you didn't originally choose.

You don't need probability theory — you need three rows. Fix your initial pick as door 1 (it changes nothing to assume this) and walk through the only three possibilities for where the car is. The host then does his one job: open a door that isn't yours and isn't the car.

Notice the engine in the middle column. When your first pick is a goat — which happens two-thirds of the time — the host has no choice: there is exactly one goat he is allowed to reveal, and opening it points straight at the car. His constraint, not your luck, is what loads the other door.

If the three-door version still feels like a trick, scale it up — a thought experiment owed to the statistician who has spent a career defending the answer. Imagine a hundred doors, one car. You pick door 1. The host, who knows where the car is, now opens ninety-eight other doors, every single one a goat, leaving just your door and one other.

Almost nobody stays in the hundred-door game. The probability your first pick was right is a flat 1%, and the host's surviving door inherits the other 99% — because he was forced to keep the car hidden if he possibly could. The three-door version is identical; it just makes the gap between one-third and two-thirds small enough for intuition to wave away.

Here is the subtlety that even careful solvers miss, and the reason Monty Hall belongs in a guide to statistical illusions. The advantage of switching does not come from a goat being revealed. It comes from who revealed it, and what they were forced to do. Change that, and the very same picture — your door, one open goat, one closed door — means something completely different.

This is the heart of it. The informed host's reveal is filtered through a rule — “never the car, never your door” — and that filter leaks information about where the car is. The clueless host's reveal carries no such information; in the worlds where he would have hit the car, the game simply doesn't reach this point, and conditioning on “he missed” leaves the two closed doors genuinely equal. Same image on the stage, opposite odds, because the process behind the image was different.

Formally this is just Bayes' theorem doing its quiet work: you must update not on what you saw but on how likely you were to see it under each possibility. It is the same discipline behind the base-rate fallacy — a piece of evidence is only as informative as the process that generated it.

1990. A reader posed the puzzle to Marilyn vos Savant's column in Parade magazine. She gave the correct answer — switch — and was buried under an estimated ten thousand letters telling her she was wrong, roughly a thousand of them from people with PhDs. “You blew it,” wrote one mathematician; “you are the goat,” offered another. She was right, and patiently said so for months until the tide turned.

The Erdős test. The great combinatorialist Paul Erdős — a man who had proved theorems most of us cannot state — reportedly refused to accept the answer until a colleague showed him a computer simulation grinding out two-thirds. If your own intuition rebelled at the top of this page, you are in distinguished company; the simulation in Figure 1 is the same cure.

The wider lesson. Monty Hall is one of a family of conditioning puzzles — the three prisoners problem, Bertrand's box, the boy-or-girl paradox — that all turn on the same hinge: evidence must be weighed by the process that produced it, not taken at face value. The trap recurs far outside game shows. Whenever information reaches you through a filter — a witness who chose what to mention, a test run only on people who were already symptomatic, a statistic someone selected because it was striking — the naive reading treats the reveal as neutral when it was anything but.

Ask that, and Monty Hall stops being a paradox and becomes a habit. The host had to avoid the car; that constraint is information; switching banks it. Most of the illusions in this compendium are, at bottom, failures to ask what the revealing process was up to.