Split-Brain
When a network partition makes two halves of a cluster each think they are in charge — risking divergent, conflicting writes.
Split-brain is a distributed-systems failure where a network partition splits a cluster into two groups that each believe the other is dead and each elect their own leader — so both accept writes, and the data diverges into conflict.
Worked example: a two-node database loses the link between nodes; each promotes itself to primary and takes writes, and when the network heals you have two conflicting histories to reconcile. Gotcha: the standard defenses are quorum (a majority must agree, so a minority partition cannot act) and fencing (a fresh leader invalidates the old one) — which is why consensus systems want an odd number of nodes: to guarantee at most one majority.