Neon uses Paxos instead of Raft for its consensus replication due to the separation of storage and compute, which allowed it to keep Postgres changes to a minimum. Neon distinguishes between several types of nodes, including compute nodes and safekeepers, and uses Paxos-like separation of concerns and recovery procedure, Raft-like election procedure, and TLA⁺ to check its modifications. The main difference between Paxos and Raft is how they are described, with Paxos having a more bottoms-up approach and Raft following a top-down approach. Neon's replication protocol is designed to solve the atomic broadcast problem with some small quirks, such as the use of separate roles for compute nodes and safekeepers. The system has a control plane that eventually resolves competing nodes, ensuring safety even in the presence of failures. Verification of the correctness of Neon's thought process was done using TLA⁺, which proved to be an excellent tool.