When proving security of the GMW protocol, we imagined an intermidiate sim S' which learns a witness. We claimed that such S' can output a view which is identically distributed as V*'s view. There was some explanation as to why this is true (conditioning on a choice of e, etc.).
My question is: assuming S' learns a witness, isn't it sufficient to claim the following general claim (not speicific to 3COL or the GMW protocol):
S' can output a view which is identically distributed as V*'s view, because it can run a full simulation of the protocol, i.e. simulate <P(w),V*>(x).