Update: I've noticed that part of the issue is it seems to be "double counting" usage from Safari.
As stated in the previous comment, the events target only category tokens, and not specific app or website tokens. Using the website x.com on Safari for 10 minutes caused the event threshold to fire 10 minutes early - so my hunch is it is counting Safari and the website x.com separately.
However, this is likely NOT the only thing that causes the overcounting, since users experience thresholds showing much higher screen time than can be only explained by their Safari usage. But I think this info can help point toward the root cause.
Also have submitted a bug report: https://feedbackassistant.apple.com/feedback/15103784