By splitting the transformation we can cache intermediate results and target enhancements (eg. adding dynamic sort) without the need to deal with static aspects of the query itself.