Might be worth adding a default threshold to voting, so that if all subsets differ more than this threshold the vector with the highest priority wins.