Yeah, it's one of those things that confused us more the more we dug into it. When we got the actual dump file, it contained something like 32 million records, and about 5 million of those were duplicates of the same userid. The duplicates were the really interesting thing to dig into: they were always the same userid, but many of them had a different email address or password in different records. That's what really convinced me of "multiple dumps, merged badly".