I'm bad at reading papers quickly so I can't be sure, but that description seems to be quite detailed, though quite sophisticated as usual for Bruneton. Btw isn't this included in the sources that he had released?
From what I got from the description, it's different to what we are doing. Our vector data are anchored to the coarsest terrain level when they should appear, and applied to the descendant tiles as post-process in the generator stage, that modifies the terrain and materials.
Terrain popping ... something I haven't bothered to implement yet, trivial things have a tendency to get pushed away
There are some potential complications when it has to play nicely with the terrain cache and the quality metrics, but a simple version should be easy to add.