There are many small and interesting details of replication behavior that don't fall under one specific section of the architecture guide, but might be useful for engineers to know. Some of these are:

• If a command errors, we still wait for write concern before returning to user
• Write concern errors don’t “undo” writes on nodes that already replicated them
• Snapshot reads vs long running majority read (where documents are read at different points in time because yielding)

We should creating a section on the architecture guide for these to be stored. We can also source more things to add from engineers outside Repl.