Good to see more leaderless consensus protocol implementations and that RAFT isn't be all and end all of all consensus problems. One advantage not mentioned in

Good to see more leaderless consensus protocol implementations and that RAFT isn't be all and end all of all consensus problems. One advantage not mentioned in the article, of leader based consensus algorithms is the ability to more easily implement read leases for faster reads. Read leases can allow for fresh reads without having to run them through the quorum protocol, by trading off availability (due to leader failure) and also correctness in certain edge cases (since read leases will depend on ability for individual machines to measure time delta with reasonable accuracy, which may not be true on some weird VM scenarios).