Overlap Is Not Enough

A team sets $N = 3$ , $R = 2$ , $W = 2$ and treats $R + W > N$ as a guarantee that a read always observes the latest acknowledged write. In which of these situations does that reasoning fail? More than one applies.

A client writes at $W = 2$ and then immediately reads at $R = 2$ through a different coordinator node than the one that handled the write.

The team grows the cluster to $N = 5$ and leaves $R$ and $W$ at 2, on the grounds that $R + W$ was already greater than the old $N$ .

Two clients write concurrently with no causal relation and the store resolves by last-write-wins, so one acknowledged write disappears.

Sloppy quorum is on, so a write can be acknowledged by fallback nodes outside the key's home replicas while a later read only consults the home replicas.

A write reaches only one replica, so it is reported as failed to the client, but the replica that did accept it never rolls it back.

One replica's disk is slower than the others, so it consistently acknowledges last.