Anomaly Detection
Looks through recent script runs for ones that stand out. The checks are plain statistics; the AI model only writes the summary, and only when something is found.
What you provide
- Script: limit the scan to one script, or leave All scripts.
- Runs to Scan: the most recent 1 to 1,000 runs; 200 when left empty. A number outside that range keeps Run Agent off and the form says why.
What it finds
| Finding | When | Severity |
|---|---|---|
| Slower than usual | A run more than 2.5 standard deviations slower than that script's average, once the script has at least 5 timed runs | Warning; error above 4 standard deviations |
| Failed after succeeding | A run that failed right after a successful run of the same script, even if other scripts ran in between | Error |
| Unusual row count | A run that changed more than 5 times, or less than a fifth of, the rows the script usually changes, once it has at least 5 runs that changed rows | Warning |
What you get
With nothing found, the answer is No anomalies in the last N runs. and the AI model is not called, so it works without an AI provider. Otherwise each finding is listed with its script and details, and the model writes a one- or two-sentence summary. When the run history cannot be read, the run fails and says why rather than reporting that nothing was found.