Currently, user profile service (UPS) maintains a map of user IDs to the experiment IDs they've previously been exposed to and the variation ID they received. The SDKs continuously append to the UPS without any cleanup, even if an experiment has been concluded and is no longer relevant.

Customers have in the past implemented diff logic to compare the live experiment IDs in the datafile vs UPS and remove experiment IDs from the UPS that are no longer in the datafile.