Monty Hall problem¶

A worked three-door puzzle in which switching wins ⅔ of the time because the host's reveal was constrained by what he knew — drilling the move of updating on the protocol that produced an observation, not on its bare surface.

Core Idea¶

The Monty Hall problem is a probability puzzle, originating with Selvin (1975, American Statistician) and given its popular form by Marilyn vos Savant in Parade magazine (1990), in which a contestant faces three closed doors, one concealing a prize and two concealing goats, picks one door, and is then shown a goat behind one of the other two doors by a host who knows the prize location and always reveals a goat. The counter-intuitive result is that switching to the remaining closed door wins the prize with probability ⅔, while staying with the original choice wins with probability only ⅓, even though exactly two closed doors remain. The mechanism is informational: the host's choice of which door to open is constrained by his knowledge — if the contestant initially chose a goat door (probability ⅔), the host is forced to open the one other goat door, leaving the prize behind the switch door; if the contestant initially chose the prize door (probability ⅓), the host can open either remaining door and the switch door has a goat. These two cases are not symmetric, and Bayesian updating against the host's constrained protocol concentrates the posterior on the switch door. The puzzle's persistent counter-intuitiveness — thousands of professional mathematicians and statisticians wrote to insist vos Savant was wrong — is documented in the cognitive psychology literature (Granberg and Brown 1995 found switching rates around 13% even after repeated trial) and makes the puzzle a canonical teaching device for the failure mode of treating post-revelation probabilities as if the protocol generating the revelation had been unconstrained.

Structural Signature¶

Sig role-phrases:

the three doors and one prize — the symmetric option set, prior ⅓ on each, that makes the naive symmetry intuition look reasonable
the knowing, constrained host — an agent who knows the prize location and whose policy forces him to open a goat door other than the contestant's pick
the contestant's initial pick — the choice that partitions the world into the ⅔ "picked-a-goat" branch and the ⅓ "picked-the-prize" branch
the protocol-constrained reveal — opening a goat door: the observation whose evidential weight is set by the host's policy, not by the bare door count
the protocol-aware update — Bayesian conditioning against the host's constraint, concentrating the posterior at ⅔ on the unopened switch door
the switch decision — the action exploiting the posterior; staying keeps the original ⅓
the policy-swap control (what it isolates) — holding doors and revealed goat fixed and replacing the knowing host with a random one collapses the posterior to ½ each, proving the information lived in the protocol
the durable wrong intuition — the documented, replication-proof default to a symmetry prior over remaining options, which makes the puzzle a teaching instrument for protocol-conditioning

What It Is Not¶

Not a genuine paradox. Popular usage calls it the "Monty Hall paradox," but the result contains no contradiction: it is a veridical result — true, provable, and uncontroversial among probabilists — that merely violates intuition. What feels paradoxical is the gap between the correct ⅔ and the symmetry reflex, not any inconsistency in the mathematics.
Not a 50/50 choice after the reveal. The tempting reading — "two doors remain, so each is equally likely" — discards the host's constraint. The reveal is not a neutral fact but a constrained act; the original pick keeps its ⅓ while the switch door inherits ⅔. Equal door count does not imply equal probability.
Not independent of the host's protocol. The ⅔ advantage depends entirely on the host knowing the prize location and always being forced to reveal a goat. Replace him with a host who opens a remaining door at random and merely happens to show a goat, and the posterior collapses to ½ each — the switching advantage vanishes. The information lives in the policy, not the door count.
Not base-rate neglect. Though both are conditional-probability failures, the Monty Hall error is specifically a failure to condition on the data-generating protocol — to update against the host's constrained choice — not a failure to weight a prior base rate. The neglected quantity is the host's policy, not a population frequency.
Not a flaw in probability theory or a trick of wording. The puzzle is not a sophistry that a sharper formalization would dissolve; the calculation is elementary and agreed-upon. The difficulty is psychological — the intuitive system defaults hard to a symmetry prior — which is exactly why it survives as a teaching device rather than a genuine open problem.

Scope of Application¶

The Monty Hall problem is a worked exemplar rather than a field-deployed mechanism, so its genuine habitats are the teaching and reasoning-science contexts where the identical three-door scenario is actually used; its reach is bounded there. (In applied inference — clinical trials, forensics, econometrics — what operates is the parent prime, protocol-aware Bayesian updating; the puzzle appears only as the case the textbook borrows, so those settings are not its habitats.)

Probability and statistics pedagogy — the home turf: the canonical worked example for conditional probability and Bayes' theorem, with its named variants (n doors, several goats revealed, the host-knowledge variants Monty Fall and Monty Crawl) themselves pedagogical mainstays from Mosteller onward.
Cognitive psychology of reasoning under uncertainty — a standard demonstration of how weakly the intuitive system conditions on the data-generating protocol, the documented durability of the wrong answer (the indignant-mathematician episode, ~13% switching after repeated play) functioning as evidence about cognition.
Decision-theory and Bayesian-inference teaching — the stock illustration that optimal choice depends on the protocol generating the data, not the data alone; it recurs in MCMC and graphical-model tutorials and AI/ML curricula to motivate modelling the data-generating process explicitly.
Analysis-as-craft shorthand — as a stored exemplar, "this is a Monty Hall situation" travels across statistics, econometrics, and forensic work as a compact label for "the process that produced the observation carries information that must be modelled, not ignored."

Clarity¶

As a teaching device the Monty Hall problem makes one reasoning failure unmistakable: treating an observation as if it had been generated unconditionally when in fact a protocol — the host's knowledge-constrained choice of which door to open — produced it. The naive move ("two doors remain, so the prize is equally likely behind each") quietly discards the host's constraint; the correct move recognizes that the reveal is not a neutral fact about the world but an act by an agent whose options depended on hidden state. The puzzle thereby sharpens the distinction between the data and the data-generating process, and shows that the same observation — "a goat is behind door 2" — carries entirely different evidential weight depending on the protocol that exposed it. Its diagnostic value is precisely that it isolates this one move: change the host to one who opens a door at random (and happens to reveal a goat) and the switching advantage vanishes, which makes vivid that the information lived in the host's policy, not in the door count.

What the puzzle additionally makes legible is the strength and durability of the faulty intuition. That trained mathematicians insisted switching could not matter, and that experimental subjects keep staying even after repeated play, demonstrates how powerfully the intuitive system defaults to a symmetry prior over the remaining options and how rarely it spontaneously conditions on the protocol. For the probability teacher this turns a single scenario into a sharp instrument: it lets a practitioner ask, of any inference, "was the observation I am updating on selected by a process whose choices depended on the answer?" — the question that separates protocol-aware Bayesian updating from the symmetry reflex it is engineered to expose.

Manages Complexity¶

The Monty Hall problem manages complexity less as a calculation than as a worked exemplar that collapses a whole class of conditional-probability mistakes into one three-door picture an analyst can carry. The unmanaged sprawl is the family of inference errors that share a single hidden structure: an observation was produced by a protocol whose options depended on the answer, yet the reasoner updates as though the observation had arrived unconditionally. That family is large and superficially heterogeneous — informative censoring in a clinical trial, the prosecutor's fallacy in court, selection-on-the-dependent-variable in econometrics, the file-drawer effect in a literature, survivorship inference from the firms that lived, the boy-girl paradox. Treated as separate puzzles, each demands its own derivation and its own re-discovery that the symmetry intuition is wrong. The puzzle compresses them by supplying a single canonical case whose answer is memorable and whose mechanism is fully exposed, so that "this is a Monty Hall situation" becomes shorthand for "the process that generated the observation carries information that must be modelled, not ignored," and the analyst recognises the structure by resemblance to one stored exemplar instead of re-analysing each instance from scratch.

What makes the exemplar do real work — rather than merely amuse — is that it isolates the single load-bearing parameter and shows the outcome swinging on it. The complexity of "how should I update on this reveal?" reduces to one question the analyst learns to ask of any inference: was the observation I am conditioning on selected by a process whose choices depended on the answer? The puzzle makes that question's stakes vivid by exhibiting both branches of the fork in the same scenario. Hold the door count fixed at two and vary only the host's policy: a host who knows the prize location and is constrained to reveal a goat concentrates the posterior on the switch door at ⅔, while a host who opens a door at random and happens to reveal a goat leaves the two remaining doors at ½ each. Identical surface — one goat revealed, two doors closed — opposite posteriors, with the entire difference living in the protocol. The analyst thus learns to read the qualitative outcome off a single binary (was assignment outcome-dependent?) rather than re-deriving the likelihood for every new setting: where the generating process is constrained by hidden state, condition on the protocol and expect the naive symmetry answer to be wrong; where it is genuinely unconstrained, the symmetry answer stands. A high-dimensional zoo of distinct-looking inference traps collapses to one diagnostic question with a two-way branch and one unforgettable case that fixes which branch is which.

Abstract Reasoning¶

The central move the puzzle drills is conditioning on the protocol, not the data — updating against the process that produced an observation rather than the observation's surface. The analyst reasons from "a goat was revealed behind door 2" not to a symmetry over the two remaining doors but to the host's constraint: he knew the prize location and was forced to reveal a goat, so the reveal is an act by an agent whose options depended on hidden state. The characteristic inference runs from the host's policy to a protocol-aware likelihood — if the contestant first picked a goat (probability ⅔) the host had no choice and the prize sits behind the switch door; if the contestant first picked the prize (probability ⅓) the host chose freely and the switch door is a goat — so Bayesian updating concentrates the posterior on the switch door at ⅔. The move is to treat which information was allowed to appear as carrying the evidential weight, not the bare fact that appeared.

The puzzle's sharpest contribution is a controlled-variation diagnostic that isolates the load-bearing variable by holding everything else fixed. The analyst reasons: keep the door count at two and the revealed goat identical, vary only the host's policy, and watch the posterior swing — a knowing, goat-constrained host gives ⅔ on the switch door, while a host who opens a door at random and merely happens to reveal a goat gives ½ each. Identical surface, opposite posteriors, the entire difference living in the protocol. This licenses the general move of asking, before crediting any reveal, what would this same observation have meant under a different generating policy? — using the contrast between the two host regimes to locate exactly where the information lives.

From that contrast the puzzle distills a single boundary-drawing question the analyst learns to put to any inference: was the observation I am conditioning on selected by a process whose choices depended on the answer? The inference is a two-way branch read off one binary — where assignment is outcome-dependent (constrained by hidden state), condition on the protocol and expect the naive symmetry answer to be wrong; where it is genuinely unconstrained, the symmetry answer stands. The move converts a zoo of distinct-looking conditional-probability traps into one decidable fork, recognized by resemblance to a single stored exemplar rather than re-derived each time.

A final, reflexive move concerns the durability of the wrong intuition itself, which the puzzle turns into evidence about cognition. Reasoning from the documented fact that trained mathematicians insisted switching could not matter and that subjects keep staying even after repeated play, the analyst infers that the intuitive system defaults hard to a symmetry prior over remaining options and only rarely conditions spontaneously on the protocol — so the failure is not ignorance of arithmetic but a representational frame that does not naturally accommodate protocol-conditioning. This licenses a predictive move for the teacher: anticipate that showing the calculation will not dissolve the resistance, and that the reliable repair is to make the host's constraint vivid (enumerate the cases, or run the random-host variant) rather than to re-present the probability.

Knowledge Transfer¶

A preliminary honesty is needed about what kind of object is transferring. The Monty Hall problem is not a mechanism deployed in the field but a worked exemplar — one perfectly calibrated three-door scenario — so its "transfer" is the transfer of a teaching device and a stored case, not of an operative apparatus. Within its home domain that transfer is nonetheless real and substantive. Across probability and statistics pedagogy it is the canonical worked example for conditional probability and Bayes' theorem, and its named variants — n doors, several goats revealed, the host-knowledge variants (Monty Fall, Monty Crawl) — are themselves pedagogical mainstays; across the cognitive psychology of reasoning under uncertainty it is a standard demonstration of how weakly the intuitive system conditions on the data-generating protocol, with the documented durability of the wrong answer (the indignant-mathematician episode, the ~13% switching rate after repeated play) functioning as evidence about cognition; across decision-theory and Bayesian-inference teaching it is the stock illustration that optimal choice depends on the protocol generating the data, not the data alone, and recurs in MCMC and graphical-model tutorials and AI/ML curricula. Across all of these the same scenario carries intact, vocabulary and all — protocol, host policy, posterior, condition on the reveal — because the substrate is the same activity, teaching protocol-aware updating, under different curricula. As a stored exemplar it also transfers within analysis-as-craft: "this is a Monty Hall situation" is usable shorthand, across statistics, econometrics, and forensics, for "the process that produced the observation carries information that must be modelled, not ignored."

Beyond the exemplar's role, what genuinely recurs across substrates is a shared abstract mechanism, and it belongs to the parent prime, not to the puzzle. The portable lesson — an observation selected by a process whose options depended on the answer must be updated against that process, not taken at surface symmetry — is protocol-aware Bayesian updating (with selection bias and the prosecutor's-fallacy / informative-censoring family as its close kin), and that mechanism is what actually operates in the substrate fields: informative-censoring corrections in clinical-trial design, non-response weighting in survey methodology, Heckman-style selection models in econometrics, selection-effect auditing of scraped corpora in AI/ML, and the prosecutor's fallacy in forensic statistics. In each of those settings the operative apparatus is the field's own — not three doors and a host — and the puzzle is present only as the thing the textbook used to make the structure memorable. So when the Monty Hall name is carried into those fields it travels as analogy and pedagogy: it lends the vivid shape of the protocol-conditioning insight while the goat-and-door machinery, the specific ⅔, and the host's policy have no referent in a clinical trial or a courtroom. The honest division is therefore twofold — within probability pedagogy the exemplar itself transfers, intact and operative-as-a-teaching-object; into applied inference the structural lesson transfers but rides on the parent prime (Bayesian updating / selection bias) it instantiates, while the puzzle remains a teaching device rather than the operative concept (see Structural Core vs. Domain Accent).

Examples¶

Canonical¶

Take vos Savant's 1990 version and enumerate. The contestant picks door 1; the prize is equally likely (⅓ each) behind doors 1, 2, or 3. Case A — prize behind door 1 (probability ⅓): the host may open door 2 or 3, and switching loses. Case B — prize behind door 2 (⅓): the host is forced to open door 3 (he cannot open the contestant's door or the prize door), so switching to door 2 wins. Case C — prize behind door 3 (⅓): the host is forced to open door 2, so switching to door 3 wins. Switching therefore wins in two of the three equally likely cases: P(win | switch) = ⅔, versus P(win | stay) = ⅓. The whole result turns on the host being forced, in Cases B and C, to reveal the one goat his knowledge permits.

Mapped back: The equal ⅓ priors are the three doors and one prize; door 1 is the contestant's initial pick splitting the world into the ⅔ goat-first branch and the ⅓ prize-first branch. The host opening a forced goat door is the protocol-constrained reveal, enumerating the cases is the protocol-aware update landing ⅔ on the unopened door, and taking it is the switch decision.

Applied / In Practice¶

During World War II, Abraham Wald and the Statistical Research Group analysed the distribution of bullet holes on aircraft returning from combat. The military's instinct was to add armour where the holes clustered — wings and fuselage. Wald pointed out that the sample was protocol-selected: these were the planes that survived. Planes hit in the engines and cockpit disproportionately did not return, so those areas showed few holes precisely because damage there was fatal. The correct inference was to armour the places with the fewest observed holes. Reinforcing the engine and cockpit areas — the "missing" holes — is the standard textbook lesson, a real inference where reading the observed pattern at face value would have armoured exactly the wrong places.

Mapped back: The returning planes are a protocol-constrained reveal: survival, like the host's knowledge, filtered which observations could appear. Armouring where the holes are is the durable wrong intuition — the surface-symmetry read. Recognising that survival depended on where a plane was hit, and inverting the map, is the protocol-aware update; imagining that all downed planes had also been inspected is the policy-swap control showing the information lived in the selection process, not the raw hole count.

Structural Tensions¶

T1: The memorable exemplar versus over-recognition by resemblance. The puzzle's whole value as a stored case is that "this is a Monty Hall situation" lets an analyst recognize a protocol-conditioning trap by resemblance to one vivid scenario instead of re-deriving each instance. But the same vividness invites false positives: a situation with two remaining options and a revealed one feels like Monty Hall even when no knowing, constrained agent produced the reveal, and the analyst imports the ⅔ where it does not hold. Recognition-by-resemblance is exactly the faculty the puzzle is meant to discipline, and it is the faculty most likely to misfire on the puzzle's own surface features. The tension is that the exemplar's power to be recognized is inseparable from its power to be mis-recognized — the crisp ⅔ is memorable precisely because it can be mislabelled onto structurally different problems. Diagnostic: Does this situation actually contain an agent whose reveal was constrained by hidden state, or only the surface pattern (two options, one shown) that makes it look like Monty Hall?

T2: Isolating the protocol versus the demand that generalizes to an unmeetable rule. The lesson is that identical surfaces ("a goat behind door 2") carry opposite evidential weight depending on the generating policy — so information lives in the protocol, not the observation. Learned as a general principle, this converts into a demand: never credit an observation without knowing the process that selected it. But in real inference the generating protocol is frequently unobservable — you rarely know the publication filter, the censoring rule, or the selection mechanism with the crispness the puzzle stipulates. The tension is that the puzzle teaches an unarguable principle whose full application is often impossible: it tells you that surface symmetry is untrustworthy without telling you how to recover the protocol you cannot see, so the honest lesson can leave the analyst more uncertain, not less. Diagnostic: Is the selecting protocol here actually knowable, or does the Monty Hall lesson only establish that the observation cannot be trusted at surface without supplying the protocol needed to reweight it?

T3: Pedagogical crispness versus the idealization that produces it. The clean ⅔ depends on strong stipulations: the host always reveals a goat, always offers a switch, and chooses uniformly when free. Those idealizations are what make the answer exact, teachable, and free of argument — and they are exactly what real problems lack, where a "host" may reveal selectively, offer switches strategically, or not exist as a single well-defined policy at all. The features that give the puzzle its didactic sharpness are the features that separate it from the applied settings it is used to illuminate. The tension is that the exemplar transfers its shape best when it is most idealized, but its idealization is precisely what must be dropped to apply it — so the cleaner the teaching case, the larger the gap between it and any real inference it is invoked to explain. Diagnostic: Do the applied situation's protocol assumptions match the puzzle's idealized stipulations, or is the crisp ⅔ an artefact of stipulations the real case violates?

T4: The durable wrong intuition as teaching asset versus as evidence the lesson does not take. The puzzle's fame rests on the resistance it provokes — indignant mathematicians, ~13% switching after repeated play — which is what makes it a dramatic instrument for exhibiting the failure to condition on protocol. But that same durability is evidence that showing the calculation does not repair the underlying frame: students who learn "switch wins ⅔" often store a fact about this puzzle rather than the generalizable move of protocol-conditioning, and the intuition reasserts itself on the next, differently-dressed instance. The tension is that the cognitive stubbornness which makes Monty Hall a compelling demonstration is the same stubbornness that limits its power to actually transfer the reasoning skill it demonstrates — it dramatizes the bug better than it fixes it. Diagnostic: Has the learner internalized the protocol-conditioning move, or only memorized the answer to the three-door case while the symmetry reflex survives intact for the next problem?

T5: The canonical telling versus the policy-swap that carries the actual lesson. The random-host variant — swap the knowing host for one who opens a door at random and happens to reveal a goat, collapsing the posterior to ½ each — is the control that proves the information lives in the protocol rather than the door count. Yet the standard presentation usually omits it, teaching only "switch = ⅔." Stripped of the contrast, the lesson degrades into a fact about doors, which is the exact misunderstanding (surface over protocol) the puzzle exists to correct. The tension is that the most common way the exemplar is taught undercuts its own point: without the policy-swap the student learns the memorable number and misses the invariant, so the canonical telling can propagate the very frame it was meant to break. Diagnostic: Was the protocol-conditioning lesson conveyed with the random-host contrast that isolates it, or only as the bare ⅔ that invites reading the result off the door count?

T6: Autonomy versus reduction (a worked exemplar or the protocol-aware-Bayesian-updating parent). The Monty Hall problem is a teaching object — one calibrated three-door scenario — not a mechanism deployed in the field, so within probability and reasoning pedagogy the exemplar itself transfers intact, vocabulary and all, and "this is a Monty Hall situation" is genuine analyst shorthand. But what actually operates in clinical trials, surveys, econometrics, and forensics is the parent prime it instantiates: protocol-aware Bayesian updating, with selection bias and the prosecutor's-fallacy / informative-censoring family as kin. Carried into those fields the puzzle travels as analogy and pedagogy — it lends the vivid shape while the goats, doors, host policy, and specific ⅔ have no referent in a courtroom. The tension is that the puzzle's fame can crowd out the parent that does the real inferential work, so people invoke "Monty Hall" where they should invoke Bayesian updating on the generating process. Diagnostic: Resolve toward protocol-aware Bayesian updating / selection bias when performing or explaining the applied inference; toward the Monty Hall problem when the goal is to teach the move with its canonical three-door case.

Structural–Framed Character¶

The Monty Hall problem is mixed on the structural–framed spectrum — its underlying content is a substrate-neutral mathematical truth realized even in observer-free selection, but the named entry is a worked pedagogical exemplar rather than a mechanism, an artifact of teaching whose door-and-goat vocabulary does not travel, so it holds the middle. The criteria pull in genuinely opposite directions, which is what makes this entry distinctive. On evaluative weight it reads structural: the puzzle convicts nothing — it is a veridical result (true, provable, uncontroversial among probabilists), not a normative judgment. Its structural core — protocol-aware Bayesian updating — is as structural as content gets: a mathematical necessity, not a human convention, and one that operates in genuinely observer-free selection processes (the Wald survivorship example is real inference where physical attrition, not a host, did the selecting), so the mechanism it teaches is not human-practice-bound at all. But the named object pulls framed: "the Monty Hall problem" is a constructed teaching device — three doors, a host, a prize — an artifact of probability pedagogy (Selvin, vos Savant), which is human-practice-bound in the sense that the scenario exists only as an instrument of instruction. Institutional origin is correspondingly mixed: the math is discovered, the puzzle is authored. Vocab-travels is low: doors, goats, the knowing host, the specific ⅔ have no referent in a clinical trial or courtroom. Import-vs-recognize is bimodal and unusually clean: within probability and reasoning pedagogy the exemplar itself transfers intact (and "this is a Monty Hall situation" is genuine analyst shorthand), while into applied inference only the parent transfers — the puzzle rides along as analogy and teaching illustration.

The portable structural skeleton is a single one: protocol-aware Bayesian updating — update on the process that selected an observation, not on its bare surface, whenever that process's options depended on the answer. That skeleton genuinely operates across clinical trials, surveys, econometrics, and forensics, but it is exactly what the Monty Hall problem instantiates from its parent prime — protocol-aware Bayesian updating, with selection_bias and the prosecutor's-fallacy / informative-censoring family as kin — not what makes "the Monty Hall problem" itself portable: the cross-domain reach belongs to that parent (whose apparatus is each field's own), while the domain-accented cargo — the three doors, the constrained host, the policy-swap control, the specific ⅔ — stays home as a teaching object, sitting alongside other pedagogical probability puzzles. Its character: an evaluatively neutral worked exemplar whose structural core is the substrate-neutral, even observer-independent mechanism of protocol-aware Bayesian updating, but whose identity as a constructed teaching device with non-traveling door-and-goat vocabulary keeps the named puzzle mixed rather than a free-floating prime.

Structural Core vs. Domain Accent¶

This section settles why the Monty Hall problem is a domain-specific abstraction and not a prime — and its case is unusual, because the entry is a worked pedagogical exemplar rather than a mechanism, so "domain" here means the teaching context, not a field of deployment.

What is skeletal (could lift toward a cross-domain prime). Strip the doors and goats away and a genuine, substrate-neutral relational structure survives: update on the process that selected an observation, not on its bare surface, whenever that process's options depended on the answer. The portable pieces are abstract — a set of outcomes, an observation whose appearance was filtered by a selecting process, a likelihood set by that process's constraint rather than by the observation's surface, and a posterior that reweights accordingly. That skeleton is as structural as content gets: a mathematical necessity that operates even in observer-free selection (physical attrition did the filtering in the Wald survivorship case, no host required), which is exactly why the entry names it as the parent prime the puzzle instantiates: protocol-aware Bayesian updating, with selection_bias and the prosecutor's-fallacy / informative-censoring family as kin. But this is the core the puzzle shares — the reason it is worth teaching — not what makes it the Monty Hall problem.

What is domain-bound. What individuates the entry is a constructed teaching apparatus that does not travel: the three doors and one prize with their symmetric ⅓ priors; the knowing, constrained host whose policy forces a goat reveal; the policy-swap control (the random-host variant that collapses the posterior to ½ and proves the information lived in the protocol); the specific answer of ⅔ on switching; and the documented durability of the wrong intuition that makes the scenario a demonstration instrument. These are the worked furniture of a probability puzzle — an authored device (Selvin, vos Savant), not a discovered fact. The decisive test: carry the entry into a clinical trial or a courtroom and the goats, the host, the door count, and the specific ⅔ have no referent; what one actually uses there is the field's own selection apparatus. Remove the teaching scenario and there is no "Monty Hall problem" left, only the parent mechanism it was built to illustrate.

Why this does not clear the prime bar. A prime's vocabulary travels and its transfer is recognition of the same mechanism, not analogy; the puzzle's transfer is bimodal and unusually clean about which part is which. Within probability and reasoning pedagogy the exemplar itself transfers intact, vocabulary and all — protocol, host policy, posterior, the named variants (n doors, Monty Fall, Monty Crawl) — because the substrate is one activity, teaching protocol-aware updating, under different curricula, and "this is a Monty Hall situation" is genuine analyst shorthand. Beyond pedagogy, into applied inference (informative censoring in trials, non-response weighting in surveys, Heckman selection in econometrics, the prosecutor's fallacy in forensics), the named puzzle travels only as analogy and illustration: it lends the vivid shape while the operative apparatus is each field's own. When the structural lesson is actually needed there, it rides on the parent prime the puzzle instantiates — protocol-aware Bayesian updating / selection_bias — not on the three-door case. So the cross-domain reach belongs to that parent; the doors, the constrained host, the policy-swap control, and the specific ⅔ are the domain-accented cargo that stays home as a teaching object, sitting alongside the other pedagogical probability puzzles.

Relationships to Other Abstractions¶

Current abstraction Monty Hall problem Domain-specific

Parents (1) — more general patterns this builds on

Monty Hall problem is a decomposition of Bayesian Updating Prime

The Monty Hall puzzle is a worked instance of Bayesian updating in which the likelihood is determined by the host's state-dependent reveal protocol.

Not to Be Confused With¶

The three-prisoners problem. The puzzle in which one of three condemned prisoners is pardoned and a warden, who knows the outcome, names a different prisoner who will be executed; the asking prisoner's pardon probability stays ⅓ while the unnamed other's rises to ⅔. This is structurally isomorphic to Monty Hall — a knowing, constrained informant whose reveal reweights the posterior — differing only in dress (prisoners/warden versus doors/host). It is the same protocol-conditioning lesson, not a distinct mechanism. Tell: is the reveal made by an agent constrained by hidden state (the same structure as Monty Hall), or is the surface merely similar?
Bertrand's box paradox. The three-boxes puzzle (two-gold, two-silver, one-of-each) where drawing one gold coin makes the other coin in that box gold with probability ⅔, not ½. It produces the identical counter-intuitive ⅔ by the identical failure — updating on the process that could have exposed the coin rather than the surface symmetry — and is often taught in the same breath. Tell: does the ⅔ arise because a selection process's options depended on the hidden answer (Bertrand's box, like Monty Hall), or from an unconstrained draw?
The boy-girl / two-child paradox. The puzzle where "a family has two children, at least one a boy" yields a ⅓ (not ½) probability of two boys — with the answer swinging on how the information was obtained (the protocol by which "at least one boy" was disclosed). It shares Monty Hall's protocol-sensitivity but has no constrained host revealing an option; the ambiguity lives in the sampling/disclosure rule rather than in a knowing agent's forced choice. Tell: is the reweighting driven by a host's constrained reveal of one option (Monty Hall), or by how a fact about a population was sampled and stated (two-child)?
Base-rate neglect. The failure to weight a prior population frequency when updating (the mammogram/disease-prevalence error). Though both are conditional-probability failures, Monty Hall is specifically a failure to condition on the data-generating protocol — the host's constrained choice — not a failure to use a base rate. The neglected quantity is the host's policy, not a prevalence. Tell: is the missing ingredient a population base rate (base-rate neglect), or the protocol that selected the observation (Monty Hall)?
Selection bias / survivorship bias. The applied-inference family in which an observed sample was filtered by an outcome-dependent process (Wald's returning aircraft, the file-drawer effect). This is Monty Hall's real-world kin — indeed the entry's Wald example is protocol-selection with physical attrition doing the host's job — but selection bias is the operative field apparatus, whereas Monty Hall is the worked teaching exemplar used to make it memorable. Tell: is this a deployed inferential correction on filtered field data (selection bias), or the three-door teaching scenario invoked to illustrate it (Monty Hall)?
Protocol-aware Bayesian updating (the parent). The substrate-neutral mechanism Monty Hall instantiates — update on the process that selected an observation, not on its bare surface, whenever that process's options depended on the answer. In clinical trials, surveys, econometrics, and forensics this parent (with selection_bias as kin) is what actually operates; Monty Hall rides along only as analogy and pedagogy. Tell: the parent (treated in a later section) is what carries the inference into applied fields; "the Monty Hall problem" as named is the three-door case for teaching the move.

Neighborhood in Abstraction Space¶

Monty Hall problem sits in a crowded region of the domain-specific corpus (35^th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Choice Paradoxes & Collective Decision-Making (14 abstractions)

Nearest neighbors

Centipede Game — 0.85
Ambiguity Aversion — 0.85
Traveler's Dilemma — 0.85
Cheap Talk — 0.85
Beauty Contest Game — 0.84

Computed from structural-signature embeddings · 2026-07-12