The current criterion is some kind of Delaunay ball, which might work OK for flat regions, but can produce super thin wedges (see issue: https://github.com/CGAL/cgal/issues/6982) when the mesh is not flat. The criterion used instead is the one used in PMP::isotropic_remeshing and PMP::remove_almost_degenerate_faces(), which is the typical angle-based surface Delaunay criterion.