I have a problem with all hydrogen models, namely some of the amino
acid residues chirality inverts in the final models.
Generally we recommend building heavy-atom-only models, as it is trivial
to add hydrogens to the final models. It's entirely possible that
hydrogens could pass "through" heavy atoms in the early parts of CG
optimization (where VDW interactions are not considered) and then be
trapped there - I suspect that's what's happening in your case. You
could certainly try playing with the optimization schedule if you really
want to build all-atom models.