I don’t know, but I can’t automatically scan projects from front-end. Maybe from back end I could periodically sample some projects, but I want to create a scoring system as well.
I do remember, however, that when I first created x-ray in v1.1 of the Web Explorer, it filtered about the same percentage of projects.
I wouldn’t call all of them spam, but in general, those are projects that I probably would not play – it filters out stuff that’s follows general patterns for art, remixes that don’t meet a certain threshold, words in the title like “announcement” or “remix” or “like for”, and more.
Exact criteria needs to be updated but it’s in the link in this post.
I like that idea too – that’s actually why I have my x-ray mechanism mark projects with 3+ hashtags to be hidden (though that can be overridden later)