One Node All Three

An internal wiki runs on a single Postgres instance. No replicas, no failover. A reviewer writes in the design doc: "CA. Consistent and available. Partition tolerance is not applicable to a single node."

Grade that claim strictly against the formal definitions (C = linearizability, A = every request received by a non-failing node returns a non-error response with no latency bound, P = the network may drop messages).

It satisfies A but not C, because one instance still interleaves concurrent transactions and can expose non-linearizable orders.

All three hold, vacuously. There is no partition that can force a choice, and the availability requirement is only imposed on nodes that have not failed, so the label says nothing about the wiki's real uptime.

It fails A. One instance has no failover, so a crash or a maintenance window makes requests fail outright.

It is really CP. A single node cut off from its clients cannot serve them, so availability is the property being sacrificed.