Internally it looks like the Fast-Quadric-Mesh-Simplification stores vertex coordinates as doubles, but the fast-simplification wrapper only loads floats.
The simplification calculations are adversely affected by lower-precision data, especially when the mesh is not "centered" (moved near the origin).
Also, for vertices that aren't collapsed, it can simplify user code for their coordinates not to be changed, and the double->single->double conversion loses information.
Internally it looks like the Fast-Quadric-Mesh-Simplification stores vertex coordinates as doubles, but the fast-simplification wrapper only loads floats.
The simplification calculations are adversely affected by lower-precision data, especially when the mesh is not "centered" (moved near the origin).
Also, for vertices that aren't collapsed, it can simplify user code for their coordinates not to be changed, and the double->single->double conversion loses information.