How you find out a flow broke

Every flow you build will break. Not might — will: an app changes a field name, an account expires, a source goes quiet, a format shifts by one character. The question this lesson answers is not how to prevent that, because you cannot, but how long it takes you to notice. That interval is the only thing you control, and it is the difference between an inconvenience and a mess.

There are two kinds of failure and only one of them announces itself. The loud kind throws an error: the connection refused, the field was missing, the run stopped. Your tool records it and, if you set it up, tells you. The quiet kind is worse — the flow runs successfully and does the wrong thing, or does nothing because its filter now matches nothing. No error, no alert, a green run history, and a sheet that stopped filling three weeks ago.

Make the loud kind reach a place you actually read. An error that lands in the automation tool's own log is not an alert; it is an archive. Route failures to wherever you would notice a message from a person — and include, in the text, which flow failed and what it was working on. "Flow error" tells you nothing at seven in the morning; "order-to-sheet failed on order 4417: field 'country' missing" tells you what to do.

Then stop the bleeding automatically. Every tool can halt a flow after N consecutive failures; almost nobody turns it on. Without it, a bad input can produce hundreds of attempts, hundreds of notifications, and — where the action sends or writes — hundreds of real consequences. Three failures in a row is a sensible stop: enough to ride out a momentary hiccup, few enough to contain a genuine break.

When a flow does break, resist the urge to fix it live. Turn it off first. A broken flow you are editing is a flow that may fire mid-edit on real data, and the second failure is usually messier than the first. Off, then reproduce with one real input, then fix, then run once by hand, then on. Four steps, two minutes, and no incident on top of the incident.

Keep one more thing: a note of what broke and why, one line per incident, wherever you keep the flow's description. After a few months the note becomes a pattern — the same source that changes twice a year, the account that expires every ninety days. Patterns get fixed properly; individual incidents just get patched, and then patched again.