The "archive command that returns 0 when the upload fails" is the one that got me — a pipeline without pipefail reported success for weeks while nothing was actually landing in the archive. I found out during a restore test, not during an incident, and that is the only reason it became a funny story instead of a bad one. Your point about restoring to a new instance instead of overwriting the damaged primary is the other half of the same habit: the restore you practice is the restore you get.