My understanding is it scrapes what it can’t meaningfully get out of an API. Public instances run into rate limiting, but private instances don’t really have that problem.