It will be ideal, at least for our use case where we trully need accurate sorting, not ballpark. But I understand it to be optional if it's too computational expensive. Anyway, I can try to reproduce it and give you a better dataset to debug in case you need it