Yes, you are right, but in that case is referring to the distribution of the Catalogue itself as a dataset itself, not to the distributions that are registered in the Catalogue i.e the ones it contains as Resources ... so that should be clearly clarified in the spec, IMHO. In my opinion a Distribution always come hand in hand with a Dataset ... i.e. there is no point in an orphaned distribution ...
Part III:
BTW another issue with Distribution in the current data model is the absence of an @id ... it is necessary in case one to refer explicitly to it as the target in a negotation
