it's not just about avoiding getting data again, because the filter search protocol kinda winds up matching stuff up sometimes, often times, unless the since/until window is decent

what it is more about is keeping a REALLY DAMN BIG database of events that you can go back and back and back and back and back... the web browsers' inbuilt event store can only store 10k events which amounts to about 512mb of data at best, and 2 months ago a reasonable estimate of all published events on nostr was near 200Gb