Skip to main content
Monitor feedback lets you tell the model which anomalies were expected, so it learns your business context and raises fewer false positives over time. title You give feedback by setting the status of an issue. There are five statuses. Only Expected changes the model. The others tell your team what is happening with the issue.

Statuses

Investigating

Use it when you are looking into the issue, so other team members do not start the same investigation. The model does not change.

Expected

Use it when the new value was expected and should not be flagged again, for example a product launch that increased signups. The anomaly is added to the training data of the model at its next update, and the thresholds adjust so similar values are not flagged.

Fixed

Use it when you fixed the underlying problem and expect the next run to pass, for example after you restored missing events in a table. The model does not change. By default the issue stays in error until the check passes again: the monitor is back within its range, or the transformation layer test passes.

No action needed

Use it when nothing has to be done and the issue will clear on the next run, for example a short delay in a load that will catch up. The model does not change.

Reset

Use it to remove a status set earlier, so the issue goes back to triage. If the issue was marked Expected, the model adjustment is undone and the monitor flags this kind of anomaly again. For example, use it when you marked an anomaly as expected and later found it was a real problem.

Setting a status

Open the issue from the monitor and pick a status: Investigating, Mark as Expected, Mark as Fixed, Mark as No Action Needed or Reset Status.

Statuses that count as resolved

In the workspace settings, Statuses that count as resolved sets which statuses resolve an issue. An issue in one of these statuses leaves triage at once, and its asset goes back to a healthy status, even if the check is still failing. title Keep Fixed out of this list, so a fixed issue stays in error until the check passes again. Use Expected to train the model, and Fixed, No action needed and Investigating to keep your team informed.