Learn it → Apply it → Save as Skill
When someone with guaranteed knowledge reveals information, your odds change, even though nothing you can see has physically moved. The instinct that says "nothing changed, so my odds didn't either" is exactly the instinct this model exists to override.
The Dr. Wallace’s Favorite Mental Models is designed to turn useful ways of thinking into reusable tools. Each model gives you a fast visual overview, deeper interactive learning, practical application exercises and an AI-ready Markdown skill you can save for future use.
Instead of merely reading about mental models, you can gradually build your own AI skills library: a collection of portable reasoning skills that can help ChatGPT, Claude, Grok and other AI assistants apply the mental models you value.
In the classic three-door version, once a host who always knows where the prize is opens a losing door on purpose, switching your pick wins two times out of three. Staying wins only one time out of three. The host's guaranteed knowledge is what makes the odds move.
Important: This only holds because the host always knows where the prize is, always opens a door you didn't pick, and always opens one that loses. Change any one of those rules and the math changes with it.
The model is most useful whenever someone (or something, including an AI system) who already knows the answer removes one or more options on purpose, and you have to decide whether that changes your odds.
Behind one door is a prize; behind the other two is nothing. Pick a door, watch the host (who always knows where the prize is) open one of the remaining doors that has nothing behind it, then choose to stay or switch.
Play several rounds choosing to stay, then several choosing to switch. The scoreboard converges toward 1-in-3 for staying and 2-in-3 for switching. This is a simple local simulation; nothing you do here is sent anywhere.
The model is a short check you run whenever a knowing party removes options from a set of choices.
Spot a moment where your options just got narrowed by someone who already knew more than you did.
Check whether the revealer was guaranteed to know the answer and constrained in what they could reveal, or whether the reveal was random or unconstrained.
If the rules match, recalculate the odds across the remaining options. Do not assume your original pick still carries its original probability.
Act on the recomputed odds, even when switching position feels like it shouldn't matter.
Open a use case to see what the model changes in practice.
If the recruiter was always going to eliminate exactly one non-viable candidate from the full pool before telling you anything, the reveal carries information about the remaining field, not just about the one eliminated candidate. Recompute rather than assuming your original favorite's odds are unchanged.
Ask whether they always intended to rule out a non-viable option before revealing terms, or whether this is a random, unconstrained statement. A guaranteed, constrained elimination should shift your confidence in the remaining choices; an arbitrary one shouldn't.
If the system reliably eliminates weaker options before presenting a shortlist, the shortlist itself carries information worth updating on. If it eliminates somewhat randomly or inconsistently, treat the narrowing as far weaker evidence.
A test that reliably and specifically rules out one possibility should shift the relative likelihood of the remaining causes, not leave them exactly as they were before the test.
If the analyst's elimination process is reliable and rule-based, the remaining scenarios deserve updated, not identical, weight. If the elimination was arbitrary or inconsistent, be skeptical of how much it should move your view.
Whenever a host or system is guaranteed to always remove exactly one known-losing option, switching your pick among the remainder is almost never worse, and is usually better, than staying put.
This is the classic illustrative example used to teach the method.
Expand for the origin of the problem and why it provoked such a strong reaction.
The puzzle's modern mathematical form traces to a 1975 letter by biostatistician Steve Selvin, published in The American Statistician, under the name "the Monty Hall problem." It is named after Monty Hall, the longtime host of the game show Let's Make a Deal, where a similar choose-a-door, reveal-a-door format appeared, though the show's actual rules and prizes varied across episodes.
It became widely known after Marilyn vos Savant answered a reader's question about it in her "Ask Marilyn" column in Parade magazine in September 1990. She correctly advised switching. The column drew an enormous volume of mail disputing her answer, including letters from readers with advanced degrees in mathematics and statistics, many of whom were initially convinced the odds were 50/50 and that she was wrong.
Yes, given the standard rules (the host always knows where the prize is, always opens a different door than your pick, and always opens one with nothing behind it). The result has been confirmed through exhaustive enumeration, simulation and formal probability proofs, and is now a standard teaching example in probability courses.
Treating the situation as if it resets to a fresh, symmetric choice between two doors once one is opened. It does not reset, because your original pick carried information forward: it had a 1-in-3 chance of being right at the moment you made it, and the host's constrained action does not change that original probability, it only concentrates the remaining 2-in-3 chance onto the one door that was not opened.
The entire result depends on the host always being constrained to open a losing door you did not pick. If the reveal could have exposed the prize by accident, the odds no longer split 2-to-1.
The two remaining doors are not symmetric. One of them carries your original 1-in-3 odds forward; the other now carries the concentrated 2-in-3 odds that used to be spread across the two doors you didn't pick.
If the host sometimes opens your door, sometimes reveals the prize by mistake, or doesn't always offer a switch, the clean 2-in-3 result no longer applies. Check the actual rules before applying the shortcut.
This model is about a single knowing party narrowing a shared set of options. It does not apply to two unrelated events that happen to occur around the same time.
Not every elimination carries the same signal. A random or careless narrowing is weak evidence; a guaranteed, rule-based narrowing is strong evidence. Confusing the two is its own mistake.
"Trust me, it's 50/50" is not a substitute for checking the actual constraint the revealer operated under. This applies to confident people and to confident AI systems equally.
the person narrowing your options does not reliably know the answer themselves, or might reveal the "winning" option by accident. Without guaranteed, constrained knowledge, the odds don't shift the same way.
the options were narrowed before you made any initial choice. The model depends on an original pick made under uncertainty, followed by a constrained reveal about the other options.
you're facing a genuinely one-shot, high-stakes decision with no chance to verify the revealer's incentives or rules. Confirm the actual mechanism before trusting the shortcut in a decision that matters.
This browser tool structures your thinking. It does not send your information anywhere or perform AI analysis.
Save as Skill converts the model into portable Markdown instructions that teach an AI when and how to apply the Monty Hall Problem, what questions to ask, what rules to verify, what mistakes to avoid and how to structure its answer.
Save the skills you find useful and organize them into your own reusable library. The Markdown is intentionally model-neutral, so it can be used wherever an AI system accepts uploaded files, project knowledge, custom instructions or reusable text context. Platform capabilities differ, so this is a portable skill file rather than a claim of one-click installation.
This is a well-known variant, sometimes called the "Monty Fall" problem. If the host has no knowledge of the prize's location and simply happens, by chance, to open a door that turns out to be empty, the odds collapse back to an even 50/50 between staying and switching. The 2-in-3 advantage for switching depends entirely on the host's guaranteed knowledge and fixed rule, not merely on a door happening to open.
With more doors, the advantage of switching generally gets larger. For example, with 100 doors where the host opens 98 known-empty doors after your pick, switching wins 99 out of 100 times, because your original pick still only had a 1-in-100 chance of being right, and the host's constrained action concentrates nearly all the remaining probability onto the single door left unopened.
The specific doors-and-goats setup is a teaching device, but the underlying lesson generalizes to real decisions: whenever someone with guaranteed, rule-bound knowledge removes options on your behalf, that removal is itself informative and should shift your odds on what remains, not leave your original assessment untouched.
The strong intuition that two remaining doors must be a fresh 50/50 choice is a deeply ingrained pattern-matching error, and it persists even in people with strong formal training, until they work through the full enumeration of cases or run a simulation themselves. It is a useful reminder that mathematical training does not make an intuition automatically reliable.