✨ Changes

Some Changes to qualifications recently, as we prepare for some big improvements to qualification propagation and batch qualifications.

Expected & False Positive have been merged into False Positive / Expected: This cleans up the slight misunderstanding around Expected where it was acting the same as False Positive but was not always used correctly. False positive / Expected should be used for any point that should be considered as a "Good point" and not alerted on in the future.

Known Error has been changed to No action needed / Known Error and the icon has been changed. This should be used to flag Known issues that require no specific action regarding resolution/investigation. These points will be ignored by the model when using Dynamic Thresholds to ensure the model does not train on anomalous values.

Reviewed has been added. This fully neutral qualification which does not impact the anomaly model should be used for a default qualification when an anomaly has been reviewed by a user who has either dealt with it or investigated it but you don't want to impact the model in any way. This is the safest qualification and feel free to use it when you like!

Quick recap of the final two who have not changed:

Fixed This will mark the point as Fixed in the interface, this qualification can be done automatically by sifflet when a point previously in anomaly is rechecked by the lookback period. As a reminder, the lookback period can be setup in incremental monitors to recheck some previous points, such as rechecking the last 7 days in pipelines where recent data can still be "patched" or "updated" a few days later!

False Negative This is the only qualification destined for non anomalous datapoints, if the model did not detect an anomaly when it should have, False Negative will guide the model to try to detect an anomaly for similar values.

What can we expect?

Reviewed and No action needed are intended target qualifications for batch qualification processes. Sifflet is planning to shortly add propagation of Incident qualifications and qualifications of monitors themselves. For example, when closing an incident you will have the ability to automatically classify all the anomalous datapoints related to the anomaly as Reviewed or No Action Needed to ensure any user looking at the monitor will know which anomalies have been dealt with !