The guidance doesn't talk about enableWhen at all - and that's the primary determiner of what gets asked and what doesn't.  There are actually two things to happen here:

1. Questionnaires should be designed with appropriate use of enableWhen such that questions are only displayed when needed.
2. CQL logic should be partitioned to be specific to groups/questions/etc. when doing so will allow it to be more efficient - though consideration should also be given to whether performing significant data gathering at the outset (even if the data is unneeded) will produce a more positive experience than intermittent data retrieval 'on demand', when such retrieval may introduce user-interface delays

Please revamp the guidance to make this interplay between enableWhen and data retrieval clearer.