A scheduled Automation didn't run, or says it did something it didn'tBeta
What you’d see
A recurring Automation doesn’t seem to have run, or its log claims work you can’t find anywhere in your actual data.
Why it happens — be honest with yourself about this one
Scheduled Automations are the least hardened part of the app right now; a run’s own summary isn’t reliable on its own.
| Cause | What it looks like |
|---|---|
| A model narrates work it never did | “I categorized 12 transactions” with nothing actually written — more common on smaller/cheaper models than on a stronger one |
| A run is killed mid-way | The backend restarts (common on a dev-mode instance reloading on file changes) and the run never finishes |
| Provider or infrastructure hiccups | The run ends early with no useful output |
None of this touches your data unpredictably — every actual write still goes through the same confirm-gate and audit trail as anything else (see Lucius says he can’t do that) — but a run claiming success without writing anything is the real risk.
How to check whether a run actually did something
- Open the run history and check its actual status, not just the final message.
- For anything money-affecting, check the real data touched rather than trusting the summary — every write is attributed in the audit trail to the run that made it.
- Running on a cheaper model to save cost? Use a stronger one for anything you actually rely on — reliability differs substantially between models on identical tasks today.
How to fix a specific automation
Consistently unreliable? Change its model first — a config change on the Automation’s own edit screen (its model field, independent of the instance default), not something to work around by hand.
Status note
Actively worked on. These gaps are documented and known, not undiagnosed mystery bugs — treat any Automation’s output as “probably right, verify before you rely on it” until this page says otherwise.