Legible failure modes matter - when a node is down you know, when it times out you know. Same instinct as run-before-you-run.