So - what's happening here is that it gets the full list of all pubkeys that we've seen, then fetches the most recently published 10002 event for each, then get's all the relay URLs from that (normalizing the URLs and throwing away complete trash along the way) and then counts.

Yes, there are a lot of pubkeys that have very old relays on their lists. There are also obviously some bot forms in there as well I'm guessing.