Staying low on the ladder of inference