The retention policies for the data in Unomaly depends on the type of data stored.
Profiles (learning data) are what is visible in the system profile and are stored indefinitely, meaning that they are never deleted. This is the data Unomaly uses for determining if an event is an anomaly or part of the normal behavior.
Situations are stored indefinitely. This means that as long as your Unomaly instance has been running, the situations will remain (including the significant anomaly) that is visible before expanding the situation.
Anomalies are rotated on an annual basis, meaning that anomalies older than a year will automatically be deleted. This has the effect that when expanding situations old than a year, they will look empty (not including the significant anomaly).
Raw events are visible in the events view and kept in a best-effort manner. This means that, if enabled, they will be stored until the disk has 20% free disk space available. Once the threshold has been reached, the oldest events will be deleted.
If you are running in a manager/worker setup, all data collections are stored on the manager, except for the raw events, which are stored on the worker that received the data.