Clipping behavior

What is the clipping behavior desired for the model element? Currently the explainer describes as:

A model element presents its three-dimensional content as though it exists inside a portal in a page, with content clipped at the z-boundary of the model element's front face, so that no content has the ability to protrude beyond the appropriate surface.

I assumed this meant clipping (discard) at the fragment level. However, when I was trying out one of the current implementations, I noticed clipping appeared to be at the mesh label. I.e. when a mesh intersected the z=0 plane of the portal surface, the entire mesh was hidden.

I originally thought this was a bug (and not what I desired), but it could be valid behavior from the description above.

What are thoughts on how clipping behavior should be defined? Should it be specified to be at the fragment level, something else, or may different UAs differ in their exact clipping behavior?

The issue names no implementation files or tests. Start by reviewing the model element explainer's clipping description and the current implementation behavior; determine whether clipping is fragment-level, mesh-level, or user-agent-dependent, then update the specification with a normative definition.