@nprofile1q... still feels a bit weird to call the G-construction a completion though, since it takes a formal object of the completion (lax diagrams in Cat) to the category that represents it