normals of the polydata are messed up after transform

[dgobbi (David Gobbi)]

That’s not a rotation matrix, it has a determinant of -1 instead of +1. How was it computed?