Automation that behaves when things break

Every automation looks good on the day it is built. What matters is the Tuesday six months later when an API is down, someone pastes a broken file, or a service quietly changes its response. This is what happens on that day.

Bad input is rejected with the reason attached, never guessed at. An outage is retried quietly before anyone is told, because most of them fix themselves in seconds. If it still fails, you get one plain alert instead of thirty identical ones. And every run is kept, so you can open any day and see exactly what happened.