When checking what preferences are saved through PrefUdate, I noticed that the number of events saved for growthexperiments-homepage-enable was 5,285 on 2020-08-17. This seemed unusually high for an option that I'd expect to be toggled by humans.
Digging further into this, I examined one of the entires and noticed that the timestamp of the event in the Data Lake was a few seconds after the user's registration timestamp. This to me indicates that the event was saved when the system set the user's Homepage preference to 1 because the user was not in the control group.
Previously, these types of events were not saved by PrefUpdate, meaning that the Growth team's process of excluding users who turned the Homepage on or off was straightforward: if a user was logged by this schema they'd changed the preference themselves. Now, we'd instead have to figure out some heuristic to remove these system events from the data. I'd prefer if we could either not log system-generated changes like these, or add some way to identify them so they can be excluded.