This book is for people who want to know what consensus algorithms are really deciding, if not a group agreement.
When a node goes quiet and you can't tell whether it crashed or is just slow, which history do you keep?
Chapter 1 starts with that fact: you can't tell a stopped node from a slow one.
If you'd rather learn the groundwork no algorithm gets to skip than memorize one more algorithm name, it's in this book.
Four chapters trace the failure model each algorithm was built to survive, from crash faults to Byzantine nodes to permissionless networks.
About 7,100 words, PDF and EPUB, one-time purchase.