So looking at the modifications to CSS, I'm a little concerned that this refers to "When the user agent is executing the painting order", since I don't think that's particularly well-defined. Different UAs might render into surfaces/layers at different times or for different reasons. If the intent is that the element be visible on screen, it might be worth saying that more explicitly. That said, maybe the cases where UAs will differ are edge cases and it's OK for the performance timeline to differ in those edge cases (e.g., which parts of an animating transformed element that's partially offscreen are rendered initially).

I think other than that (plus the issues discussed in #326), we're fine with this, so I think it makes sense to close this review.

(Feel free to ping me at some point if you want me to have another look at the CSS integration stuff if you change it.)