Comment by nuancebydefault

I think "model welfare" is just a generalisation of "model behaving in a sane way", which is the real goal.