It’s on the contrary very simple: Lionel’s interview video about a month ago made it very clear now.
In short: they are reducing the poly draw count in directly changing the way the renderer is “using” index buffer. In other words and per my understanding they seem to tell the video card to “skip” every other triangle by virtue of “skipping” the indices to vertices for example. No matter how you do, just “skipping” vertices is guaranteed to shape morph the 3D model.
In order to do it in a way which is visually better, you not only have to “skip” vertices (which you can do via the indices) but you also have to re-adjust the vertices to preserve the overall shape. This is more complex and more computing than just skipping indices, but it is also more visually correct.
 Here is a sample video:
From an old paper:
View-dependent refinement of progressive meshes
And a few others found on their GitHub page:
Smooth view-dependent level-of-detail control and its application to terrain rendering
PS: It is published by Hugues Hoppe and he was at the time of the publication from Microsoft Research…