Not sure I like this name. Can we pick something that isn't a reference to ::part()?

Yup, would love a different name, but having a problem coming up with one. Problem is coming up with something that's not just a synonym of "abiding". We could try to skirt a little farther in the "Tree-X" namespace - tree-located? tree-positioned? tree-structural?- to indicate that not only do they abide by the tree structure generally (insofar as having a parent/child relationship) but actually have a definite position in that tree (and thus know their sibling relationships too).

Or we could go further afield. Semi-real? Pseudo-real?

"Tree-structural" is actually really grabbing me, given its parallel with the tree-structural pseudo-classes.

Well ::marker has a definite position in the tree, but you're not including it in this set of things afaict. So it's not about it's tree position, it's about the fact that generates a box that behaves exactly like any other element's generated box.

I was under the impression that the reason ::marker didn't act like ::before/etc is because we specifically hadn't nailed down its exact tree position. If it does have an exact position, then what is the reasoning behind limiting it? (Besides applying a few one-offs, like disallowing it from being display: list-item or something; that's fine for this new category of pseudos to do.)

This isn't helping to illustrate pseudo-elements that "interact with some other platform features as if they were real elements". It's illustrating a real element acting like a pseudo-element (by being selected by one), which is kindof the opposite!

I think you're misunderstanding the example, because it illustrates the exact opposite. ::part(preceding-text) is selecting a pseudo-element (the p::before pseudo).