Consistency, Availability, Partition Tolerance
The idea is that you pick 2 out of the three options. This is a bit of a misnomer since network partitions are a kind of fault that happen whether you like it or not.
Consistency
Concerned with coordinating the state of replicas in the face of delays and faults. Strong consistency is achieved using Linearity. The combination of Linearity and serializability in Software Transactions is called struct serializability or strong one-copy serializability.
When a network fault occurs, you have to choose between linearizability or total availability. Therefore, a better way to phrase this is consistent or available when partitioned.
Basically, the idea here is that if your application requires linearizability and some replicas are disconnected, you need to wait until the partition resolves or return an error. Otherwise, if linearizability is not required, the replicas can return data but it may be stale and not linearizable.