Raft makes a cluster agree on a replicated log by electing one leader per term. A candidate wins only with votes from a majority — ⌊N/2⌋+1 — which guarantees no two leaders in the same term (any two majorities overlap in at least one node). The leader appends entries and an entry commits once it is replicated on a majority. Majority everywhere is the whole trick: it makes split-brain arithmetically impossible.
Pick a cluster size and watch the majority threshold decide everything — a candidate with a majority of votes becomes leader, an entry on a majority of nodes commits, and anything short of a majority does neither: live demo
“Distributed consensus is inherently too subtle to follow.” — that reputation belonged to Paxos. Raft was designed for understandability: separate leader election, log replication, and safety into parts a person can hold in their head, and lean on one idea — the majority quorum — for correctness. cited
Why a majority and not just “enough”? Because any two majorities of the same set must share a member, and that shared node refuses to vote twice or hold two truths — so two conflicting decisions can never both stand. Consensus falls out of set arithmetic. Raft 2014
On the canonical compiler, for N=5 the majority is 3: three votes elect a leader, three replicas commit an entry, and two of either does neither: