Skip to content

Common Knowledge

Prime #
713
Origin domain
Game Theory And Strategy
Subdomain
epistemic foundations → Game Theory And Strategy
Aliases
Common Knowledge of Rationality

Core Idea

A fact p is common knowledge among a group when every member knows p, every member knows that every member knows p, every member knows that, and so on recursively to arbitrary depth. Common knowledge is the infinite-tower limit of shared knowledge: ordinary "we both know" sits one or two rungs up the tower, and the full tower is a strictly stronger condition. The structural significance is that many coordination outcomes, conventions, and equilibria are achievable only when the underlying enabling fact reaches the top of the tower, and become unachievable the moment a perturbation severs even one rung. Common knowledge is the structural pivot that converts private information into a shared epistemic substrate on which collective action can ride.

The load-bearing structure is a hierarchy of nested belief operators — K(p), K(K(p)), K(K(K(p))), and so on — together with a threshold operation that lifts the hierarchy to its infinite limit, typically a public, simultaneous, witnessed event in which everyone sees everyone else seeing the same thing. The condition is fragile in a specific way: each rung depends on every message reaching every recipient without ambiguity, so over an unreliable channel the tower cannot be completed by any finite number of acknowledgements. This is the content of the Coordinated Attack / two-generals result, where no finite exchange ever upgrades mutual to common knowledge. The prime is defined in purely formal terms — agents, a fact, nested knowledge operators, a lifting event — with no commitment to any medium, which is why it applies identically to humans, institutions, and machines.

How would you explain it like I'm…

Everybody Knows Everybody Knows

Common knowledge is when everybody knows something, and everybody knows that everybody knows it, on and on. Think of when a teacher says a rule out loud to the whole class at once. Now nobody can say 'I didn't know,' and everyone knows nobody can say that. Seeing it happen together, in the open, is what makes it stick.

The Knowing Tower

A fact is common knowledge in a group when everyone knows it, everyone knows that everyone knows it, everyone knows that, and so on forever. It's stronger than just 'we both know' — that only goes up a step or two, but common knowledge goes all the way up. The way you usually get it is a public, all-at-once event where everyone sees everyone else seeing the same thing, like an announcement in front of the whole room. A lot of teamwork and agreements only work when a fact reaches the very top of this tower. And it's fragile: cut just one rung and the whole thing can fail.

The Infinite Knowing Tower

A fact p is Common Knowledge in a group when every member knows p, every member knows that every member knows p, and so on recursively to any depth. It's the infinite-tower limit of shared knowledge: ordinary 'we both know' sits one or two rungs up, while the full tower is strictly stronger. This matters because many coordination outcomes, conventions, and equilibria are achievable only when the enabling fact reaches the top of the tower — and they collapse the moment one rung is severed. The structure is a hierarchy of nested 'knows' operators — K(p), K(K(p)), and so on — plus a lifting operation that reaches the limit, usually a public, simultaneous, witnessed event where everyone sees everyone else seeing the same thing. It's fragile in a specific way: over an unreliable channel no finite number of acknowledgements can complete the tower, which is the heart of the two-generals result.

 

A fact p is Common Knowledge among a group when every member knows p, every member knows that every member knows p, every member knows that, and so on recursively to arbitrary depth. Common knowledge is the infinite-tower limit of shared knowledge: ordinary 'we both know' sits one or two rungs up the tower, and the full tower is a strictly stronger condition. The structural significance is that many coordination outcomes, conventions, and equilibria are achievable only when the underlying enabling fact reaches the top of the tower, and become unachievable the moment a perturbation severs even one rung. Common knowledge is the structural pivot that converts private information into a shared epistemic substrate on which collective action can ride. The load-bearing structure is a hierarchy of nested belief operators — K(p), K(K(p)), K(K(K(p))) — together with a threshold operation that lifts the hierarchy to its infinite limit, typically a public, simultaneous, witnessed event in which everyone sees everyone else seeing the same thing. The condition is fragile in a specific way: each rung depends on every message reaching every recipient without ambiguity, so over an unreliable channel the tower cannot be completed by any finite number of acknowledgements — the content of the Coordinated Attack / two-generals result. The prime is defined in purely formal terms — agents, a fact, nested knowledge operators, a lifting event — with no commitment to any medium, so it applies identically to humans, institutions, and machines.

Structural Signature

the group of agents — the target fact — the tower of nested knowledge operators — the infinite-limit threshold (the lift to the top rung) — the lifting event (public, simultaneous, witnessed) — the channel-reliability fragility — the qualitative gap between mutual and common knowledge

The pattern is present when each of the following holds:

  • A group of agents. Two or more parties — people, institutions, machines — whose joint action depends on what each believes about the others.
  • A target fact. A proposition p whose epistemic status among the group is decision-relevant for some coordination, convention, or equilibrium.
  • A tower of nested knowledge operators. The structural object: everyone knows p, everyone knows that everyone knows p, and so on — K(p), K(K(p)), K(K(K(p))) — to arbitrary depth.
  • An infinite-limit threshold. Common knowledge is the top of the tower, qualitatively stronger than mutual knowledge (one or two rungs); the relevant coordination outcomes become achievable only at this limit.
  • A lifting event. A public, simultaneous, witnessed occurrence — announcement, ritual, broadcast, shared display — in which everyone sees everyone else seeing the same thing, raising p to the top rung in one stroke.
  • A channel-reliability fragility. Each rung depends on messages reaching every recipient unambiguously; over a lossy channel no finite exchange of acknowledgements completes the tower (the two-generals / coordinated-attack impossibility), so a single severed rung collapses the condition.

Composed, these make coordination a question not of who holds the information but of how high in the belief hierarchy it has been lifted, and which event performs (or blocks) that lift.

What It Is Not

  • Not coordination. Coordination is the outcome — agents acting together; common knowledge is the epistemic precondition that makes certain coordination outcomes reachable. The tower is what enables coordination, not the joint action itself.
  • Not consensus_problem. Consensus is the decision- procedure question of how to act together given the limit is unreachable; common knowledge is the epistemic limit itself. Consensus protocols exist precisely because exact common knowledge is unattainable over lossy channels (see consensus_problem).
  • Not shared_mental_model or common_ground. Those name overlapping content between agents — what they jointly represent. Common knowledge is the strictly stronger recursive condition (everyone knows that everyone knows…), not mere first-order overlap.
  • Not wisdom_of_the_crowds. That concerns aggregating independent private judgments into an accurate estimate; common knowledge concerns recursively shared belief about a single fact, with no aggregation and no accuracy claim.
  • Not reflexivity_self_reference. Reflexivity is a single system taking itself as object; the common-knowledge tower is many agents nesting beliefs about each other's beliefs — interpersonal recursion, not self-application.
  • Common misclassification. Calling mutual knowledge ("everyone knows") common knowledge. Coordination can fail despite everyone privately knowing the fact, because the upper rungs (everyone knows that everyone knows) were never lifted — the diagnostic is a missing rung, not missing information.

Broad Use

  • Game theory and economics — common knowledge underwrites Nash equilibrium reasoning (each player knows others' rationality and that others know it), rational-expectations equilibria, and the no-trade and agreement theorems.
  • Distributed systems — the impossibility of achieving common knowledge over a lossy channel (the Coordinated Attack result) drives the entire design space of consensus protocols that only approximate common knowledge under bounded assumptions.
  • Linguistics and pragmatics — shared context, conventions, and conversational presupposition rest on what speakers can take to be common ground; reference, metaphor, and irony all assume specific common-knowledge scaffolds.
  • Social norms and conventions — a norm becomes binding only once compliance is common knowledge, which is why public rituals (parades, anthems, televised events) generate it: everyone sees everyone else seeing the same thing.
  • Revolutions and collective action — mass mobilization cascades after a single visible breach of a regime's appearance of invincibility, because the breach lifts shared dissent onto the top rung.
  • Cryptography — zero-knowledge proofs, authenticated broadcast, and Byzantine agreement are defined in terms of what becomes common knowledge among honest parties without leaking to adversaries.
  • Cognitive science — human social cognition routinely operates at depths of two to four levels of recursive belief attribution, a structure shared with some non-human animals.

Across these the substrate ranges from formal games to street protests to replicated databases, while the structural object — the recursive belief tower and the event that lifts a fact to its top — is invariant.

Clarity

The prime separates three states that ordinary talk conflates: private knowledge (I know p), mutual knowledge (everyone knows p), and common knowledge (the whole recursive tower). The difference between the second and third is qualitative, not a matter of degree, and naming it makes a whole class of coordination phenomena legible. The canonical dramatization is the blue-eyes puzzle, in which publicly announcing "at least one of you has blue eyes" to a group where everyone already privately knows it nonetheless triggers an otherwise-impossible chain of inference — because the announcement upgrades mutual to common knowledge, and the new top rungs are decision-relevant.

The clarity is also diagnostic on the failure side. Saying "this isn't common knowledge, it's just mutual knowledge" identifies a structurally distinct regime in which coordination will fail despite everyone privately knowing the answer. This relocates the explanation of a stuck coordination from the participants' competence or will to a missing upper rung: someone is unsure whether others know that they know. Naming the gap between mutual and common knowledge thus converts "they all know it, why can't they act?" from a paradox into a precise structural diagnosis.

Manages Complexity

The prime collapses a potentially infinite hierarchy of nested-belief states into a single named threshold. Once something is labeled "common knowledge among the group," one can reason about its consequences without re-deriving each meta-level, treating the whole tower as a single object. The complementary collapse on the failure side is equally useful: identifying a state as merely mutual knowledge flags a distinct regime with its own predictable behavior, so the analyst need not track the rungs individually to know that coordination is unavailable.

This compression turns a family of otherwise-separate problems into instances of one question. Whether the setting is a cryptographic protocol, an organizational announcement, a social movement, or a software commit, the operative issue is the same: which facts must reach the top of the tower, and what event lifts them there? Managing the complexity of multi-agent coordination reduces to managing the elevation of specific facts through the belief hierarchy, and to recognizing which channel reliabilities and which public devices can or cannot complete the lift.

Abstract Reasoning

The prime lets a reasoner ask, of any collective-action setting: What facts must reach common knowledge for this equilibrium, norm, convention, or protocol to hold? By what channel does information climb from private to mutual to common knowledge? What perturbation would break that climb — lossy messages, private channels, deniability? What designed device — public announcement, ritual, broadcast, shared display — raises a fact onto the top rung? These questions transfer wholesale between cryptographic protocol design, organizational announcement strategy, social-movement mobilization, and software consensus.

The reasoning generalizes because the belief tower is medium-neutral. The same analysis that explains why a televised protest can flip a stable autocracy explains why a quorum acknowledgement lets a replicated database advance: in both cases what changed is the recursive depth of shared knowledge, not the underlying fact. The reasoning habit the prime installs is to treat coordination not as a matter of who holds which information but as a matter of how high in the belief hierarchy that information has been lifted, and to look for the specific event — or its absence — that performs the lift.

Knowledge Transfer

The prime carries portable interventions keyed to manufacturing or denying common knowledge. Manufacture common knowledge via public displays — all-hands meetings, signed treaties, televised ceremonies, shareholder filings — devices that ensure everyone sees everyone seeing, useful whenever a norm or commitment must bind. Deny common knowledge to preserve deniability — off-the-record briefings, backchannel diplomacy, leak-only signals — devices that propagate information privately while blocking the recursive tower. Design protocols to approximate common knowledge under bounded faults — timeout-based and quorum-based agreement, with explicit recognition that exact common knowledge is unreachable over lossy channels but adequate approximation is cheap. Diagnose coordination failures by looking for a missing upper rung when a group all knows something yet cannot act. And use ambiguity strategically — leaders deliberately keep policies vague to prevent the formation of common knowledge that would crystallize opposition.

The transfer holds because the object underneath — a set of agents, a fact, a tower of nested knowledge operators, and an event that lifts the fact to the tower's limit — is identical whether the agents are citizens, replicas, negotiators, or honest cryptographic parties. A communications team staging a public commitment, a protocol designer building quorum acknowledgement, and an organizer broadcasting visible dissent are doing the same structural work: elevate a target fact to the top of the belief hierarchy through a public, witnessed, simultaneous event, or, conversely, route information so that the elevation is blocked. The prime is fully structural — defined mathematically, carrying no home vocabulary or evaluative weight — so its diagnostics and devices recognize the same pattern across game theory, distributed systems, social movements, and cryptography without any need for translation.

Examples

Formal/abstract

The blue-eyes (muddy-children) puzzle is the prime's mechanism isolated to a single inference chain. A group of perfect logicians each has either blue or brown eyes; each can see the others' eyes but not their own, and the convention is that anyone who deduces they have blue eyes must leave at midnight. Suppose \(n\) of them have blue eyes. Before any announcement, "at least one person has blue eyes" is already mutual knowledge — with \(n \ge 2\), everyone sees a blue-eyed person — yet no one ever leaves. A visitor then publicly states, "at least one of you has blue eyes," and on the \(n\)-th midnight all \(n\) blue-eyed logicians leave simultaneously. The puzzle's force is that the announcement told no one a fact they did not already privately know; what it added was the lifting event that raised the fact to the top of the nested knowledge tower. With \(n=2\), Alice reasons: "if I have brown eyes, Bob sees no blue-eyed person, so the announcement tells Bob he is blue and he leaves the first night"; when Bob does not leave, Alice deduces her own eyes are blue. The induction needs not just that everyone knows the premise, but that everyone knows that everyone knows it, to depth \(n\) — exactly the rungs the public announcement supplies and that mutual knowledge alone does not. The channel-reliability fragility is implicit: only a public, simultaneous, witnessed statement completes the tower; a private whisper to each child would never trigger the cascade.

Mapped back: The blue-eyes puzzle instantiates every role — the logicians as the agent group, "someone has blue eyes" as the target fact, the depth-\(n\) inference as the nested-knowledge tower, the visitor's public statement as the lifting event, and the failure of any private telling to reproduce the result as the witnessed-event requirement.

Applied/industry

Distributed-database commit protocols and authoritarian-regime collapse are the same structure in machine and social substrates. In a replicated database, a coordinator cannot make replicas agree to commit a transaction merely by sending each one a "commit" message: over an unreliable network, a replica that receives the message does not know whether the others received theirs, so it cannot safely act — the channel-reliability fragility and the Two Generals impossibility, which prove that exact common knowledge is unreachable over a lossy channel. Protocols therefore approximate the lift with quorum acknowledgement: a replica acts once it knows a majority has acknowledged, which is the engineered substitute for the unreachable top rung, sufficient under bounded fault assumptions. The social analogue is sharp: an autocracy can remain stable while everyone privately resents it, because each citizen, unsure whether others will act, dares not move — dissent is mutual but not common knowledge, the missing upper rung that explains "they all know it, why can't they act?". A single visible lifting event — a televised mass protest, a viral video of a regime breach, a public square filling — raises shared dissent to the top of the tower, and mobilization cascades because now everyone knows that everyone knows that everyone will act. The corresponding interventions are duals: regimes deny common knowledge by censoring public gatherings and broadcasts (keep dissent private), while organizers manufacture it through visible, simultaneous, witnessed displays.

Mapped back: Database commit and regime collapse realize the prime end-to-end — replicas/citizens as agents, "commit" or "the regime is vulnerable" as the target fact, the unreliable network or fear of isolation as the channel fragility blocking the tower, and quorum acknowledgement or a public protest as the lifting event (or its censorship as the blocked lift).

Structural Tensions

T1 — Exact common knowledge versus achievable approximation (boundary). The infinite tower is the condition coordination wants, but the Two Generals result proves it is unreachable over any lossy channel, so every real system runs on a finite approximation. The failure mode is reasoning as if exact common knowledge had been achieved — designing a protocol or a commitment that assumes the top rung when only a few rungs were ever secured. Diagnostic: ask whether the channel is perfectly reliable; if not, the system holds bounded mutual knowledge, and any guarantee that depends on the true limit (rather than a quorum approximation) is unsound.

T2 — Lifting event versus its fragility (temporal). A public, simultaneous, witnessed event lifts a fact to the top rung in one stroke — but the same event-dependence makes the state perishable: a single ambiguity, a person who blinked, a participant who joined late, and the lift is incomplete. The failure mode is assuming a one-time announcement created durable common knowledge when partial reach or fading memory has already severed a rung. Diagnostic: ask whether every relevant agent witnessed every other witnessing; if reach was partial or the moment has passed without reinforcement, the tower has gaps the original event no longer covers.

T3 — Manufacturing versus denying (sign/direction). The prime's two intervention families point in opposite directions — public displays create common knowledge, backchannels and ambiguity deny it — and the same actor often needs both, which can collide. The failure mode is manufacturing common knowledge of a commitment one later needs deniability on, or maintaining strategic ambiguity that blocks the coordination one actually needs. Diagnostic: ask which coordination outcome the fact must enable or prevent; binding a norm and preserving deniability are mutually exclusive for the same fact, and conflating them produces a fact that is neither safely public nor usefully private.

T4 — Coordination value versus crystallized opposition (sign). Lifting a fact to common knowledge enables collective action — which is a benefit when you want the coordination and a hazard when the coordinating group is your opposition. The failure mode is a leader or platform broadcasting a fact for one purpose (transparency, announcement) and thereby manufacturing the common knowledge that crystallizes dissent or a bank run. Diagnostic: ask who gains the ability to coordinate once this fact reaches the top rung; the same lift that aligns allies aligns adversaries, and a public device is indifferent to which group it empowers.

T5 — Depth required versus depth attainable (scalar). The theory invokes an infinite tower, but real cognition operates at two-to-four levels of recursive belief, and protocols approximate only a few rungs. Most coordination needs only finite depth — yet some outcomes genuinely require more rungs than the agents can represent. The failure mode is assuming the needed depth is available when the agents (human or machine) cannot actually track "everyone knows that everyone knows that…" to the required level. Diagnostic: ask how many rungs the coordination actually requires versus how many the agents can hold; where the requirement exceeds cognitive or protocol depth, common-knowledge reasoning predicts coordination the substrate cannot in fact sustain.

T6 — Common knowledge of the fact versus of its meaning (scopal). The tower can be completed for a proposition while agents still differ on what it implies for action — everyone knowing "the signal fired" is not everyone agreeing on what to do about it. The failure mode is treating common knowledge of a fact as if it delivered common knowledge of the shared plan, when the recursive agreement covers the observation but not the interpretation or the equilibrium to play. Diagnostic: ask whether what reached the top rung is the bare fact or the action it should trigger; coordination needs the latter, and a perfectly witnessed event can leave the response ambiguous, so the lift is necessary but not sufficient.

Structural–Framed Character

Common knowledge sits at the structural pole of the structural–framed spectrum: it is a pure epistemic-formal object — the infinite tower of nested knowledge operators K(p), K(K(p)), K(K(K(p))), lifted to its limit — and its frontmatter grade (label structural, aggregate 0.0, all five criteria zero) records that every diagnostic points one way.

Walk them. The pattern carries no home vocabulary that must travel with it: the same recursive belief tower is told in the game theorist's equilibrium reasoning, the distributed-systems engineer's quorum acknowledgement, the linguist's conversational presupposition, the organizer's public protest, and the cryptographer's authenticated broadcast — each in its own words, because the structural object (a fact lifted to the top of the nesting hierarchy by a public, witnessed, simultaneous event) is medium-neutral. It carries no evaluative weight: completing the tower is neither good nor bad until you specify what coordination it enables — the same lift aligns allies or crystallizes a bank run (the prime's own T4). Its origin is formal in the strongest sense — agents, a fact, nested knowledge operators, and a lifting event, defined mathematically with no appeal to human norms; the two-generals impossibility is a theorem, not an institution. It is not human-practice-bound: the identical tower governs a televised revolution, a replicated database advancing a commit, and the recursive belief attribution observed in some non-human animals, with no human practice required for the structure to obtain. And invoking it merely recognizes the recursive depth a fact has (or has not) reached — it imports no interpretive frame, only the observation of which rung the fact sits on and which event performs the lift. On every diagnostic, it reads structural.

Substrate Independence

Common knowledge is fully substrate-independent — composite 5 / 5 on the substrate-independence scale. It is a pure epistemic-formal object — the infinite tower of nested knowledge operators K(p), K(K(p)), … lifted to its limit by a public, simultaneous, witnessed event — defined mathematically with no commitment to any medium, so it is recognized rather than translated. Domain breadth is maximal: the recursive belief tower underwrites Nash-equilibrium and no-trade reasoning in game theory and economics, the Coordinated-Attack impossibility driving consensus protocols in distributed systems, conversational presupposition in linguistics, norm-binding public rituals in social dynamics, mobilization cascades in collective action, authenticated-broadcast definitions in cryptography, and recursive belief attribution in human and some non-human animal cognition. Structural abstraction is total: agents, a fact, nested knowledge operators, and a lifting event compose a medium-neutral object, with the two-generals result holding as a theorem rather than an institution. And transfer evidence is heavily documented through carriers that port intact — the blue-eyes induction, the Two Generals impossibility, and quorum acknowledgement as the engineered substitute for the unreachable top rung are the same structure whether the agents are citizens, replicas, or honest cryptographic parties. Maximal on every component, it is a canonical 5.

  • Composite substrate independence — 5 / 5
  • Domain breadth — 5 / 5
  • Structural abstraction — 5 / 5
  • Transfer evidence — 5 / 5

Neighborhood in Abstraction Space

Common Knowledge sits in a sparse region of abstraction space (63rd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely rather than landing on a neighbor.

Family — Uncertainty, Risk & Proxy Distortion (22 primes)

Nearest neighbors

Computed from structural-signature embeddings · 2026-06-14

Not to Be Confused With

The deepest confusion is with coordination, the prime's nearest embedding neighbor, because the two are so tightly coupled that they are often treated as one. Coordination is the outcome: agents successfully acting together toward a joint equilibrium, convention, or plan. Common knowledge is the epistemic substrate that makes a class of coordination outcomes reachable at all — the recursive belief tower without which agents, each unsure whether the others know, dare not move. The distinction is exactly what explains the prime's signature diagnostic: a group can fail to coordinate despite everyone privately knowing the answer, because the explanation lives not in the agents' competence or will but in a missing upper rung of the belief hierarchy. An analyst who conflates the two will look for the failure in the wrong place — in incentives or capability — when it is epistemic. Common knowledge is necessary (often) but not sufficient for coordination; the lift can be perfect and the action still ambiguous (the prime's own tension T6).

A second genuine confusion is with consensus_problem, which the prime's own Core Idea calls its structural inverse. Common knowledge is the epistemic limit — the infinite tower of nested belief — and the Two Generals result proves that limit is unreachable over any lossy channel. The consensus problem is the decision-procedure response: given that exact common knowledge cannot be attained, how should distributed agents nonetheless agree to act, and what relaxation (synchrony, determinism, fault-tolerance) must they accept? Consensus protocols like Paxos and Raft are engineered substitutes for the unreachable top rung, approximating it with quorum acknowledgement under bounded assumptions. The two are duals: common knowledge names what coordination wants and proves it impossible; consensus names what is buildable in its absence. Confusing them leads to the error of designing a protocol that assumes the limit was achieved when only a finite approximation was ever secured — the prime's tension T1.

A third confusion worth marking is with shared_mental_model (and the candidate common_ground). These name overlapping representational content — what two agents both hold in mind, what context they jointly presuppose. Common knowledge is the strictly stronger and structurally different recursive condition: not merely that we share a representation, but that each of us knows the other knows it, and knows that they know, without bound. A shared mental model can be entirely first-order — both parties picture the same plan — and still lack the upper rungs that common knowledge requires, which is why a fact can be in everyone's head (mutual knowledge, shared model) yet not be common knowledge. Reference, irony, and convention rest on the recursive scaffold, not on mere content overlap; treating a shared mental model as common knowledge over-credits first-order agreement with a recursive property it does not have.

For a practitioner these distinctions are operational. If the problem is a failed joint action, ask whether the obstacle is epistemic (a missing rung — common_knowledge), procedural (no agreement mechanism under faults — consensus_problem), or representational (no shared model at all). Each diagnosis points to a different intervention: manufacture common knowledge via a public witnessed event, build a quorum-based protocol, or first establish shared content. The prime's unique contribution is the recursive belief tower and the lifting event that completes it — a structure none of its neighbors supplies.

Solution Archetypes

No catalogued solution archetypes reference this prime yet.