I am not sure where the idea about perfect twinning not working came from—is there a source for that?

I thought, also, that one can really do twin refinement down to ~5%. It’s only one parameter more, no? And if you really have that twin fraction, why not model it?

JPK