A theorem in distributed systems stating that, in the presence of network partitions, a distributed data store can provide at most two of these three guarantees simultaneously: consistency (all nodes see the same data at the same time), availability (every request receives a response), and partition tolerance (the system continues to operate despite message loss or network splits).

Principle

Under asynchronous networks where partitions can occur, designers must trade off between consistency and availability when a partition exists; one cannot simultaneously guarantee all three properties in all failure scenarios.

Demonstration

If a network split separates a cluster into two halves, a replicated key-value store that remains available on both sides may return divergent values (favoring availability and partition tolerance), whereas a store that enforces strict consistency may reject requests on one side until the partition heals (favoring consistency and partition tolerance).

Misapplication

Interpreting CAP as strictly forbidding any system from providing both consistency and availability ever; or conflating CAP categories with latency, durability, or nuanced consistency models like eventual or causal consistency.

Consequence

Guides architecture choices for distributed systems: under partition risk, choose CP (consistency+partition tolerance) to prioritize correctness, or AP (availability+partition tolerance) to prioritize serving requests, while CA is only possible when partitions cannot occur or are ignored.

Reversal

In a perfectly reliable network with no partitions, a system can be both consistent and available (CA), but real networks are fallible, so partition tolerance must be considered in practical designs.

Boundary

Formal statement applies to the model allowing network partitions and to binary notions of consistency and availability; it does not quantify degrees of consistency, nor prescribe specific mechanisms—practical systems use relaxed consistency, partial availability, retries, and other techniques outside the binary CAP classification.

Semantic Tension

Tension arises between the coarse CAP trichotomy and richer modern consistency taxonomies (e.g., eventual, causal, strong consistency) and between theoretical worst-case partition events and typical operational behavior; designers must map CAP choices to application-level requirements and observable trade-offs.

Synthesis

The CAP theorem formalizes a fundamental trade-off in distributed system design: when partitions are possible, you cannot guarantee consistency, availability, and partition tolerance all at once, so architects must choose which guarantees to prioritize or adopt intermediate/relaxed models that balance them in practice.