The feedback I am getting suggests that the truth of the matter is that NGB + global choice is an extension of NGB + set choice that is conservative for the language of pure sets. Can anyone confirm this, preferably with a reference? thanks Thomas