The reason I think the quality and quantity dropped was the internal schedules barely (or don't) leave enough time for the engineering work so there's very little time available for high level documentation. Internally tons of "documentation" existed as Radar comments or exchanges on internal mailing lists. Maybe a group's wiki had some crystallized documentation or high level architectural descriptions but good luck accessing it from outside that org. My favorite was some discussion about overall design or architecture that got the "let's take this offline" where all the helpful details ended up shared in an in-person meeting.
The internal secrecy and rapid development pace made it really difficult to get good overviews of technologies internally. I really sympathize with outside developers trying to cobble together an understanding of something where the documentation sucks or is missing.
The DocC tooling isn't going to be anymore effective than doxygen or other auto generated documentation without good architectural documentation. A function definition is nice but knowing you need to decombulate a frobnob before snizzlizing or that decombulation can only safely be done on the main thread is often more important. I can read a function signature but I can't necessarily know all the keys that go into some configuration NSDict passed into it.
I hope the situation improves if the DocC tooling lowers the friction for writing documentation. It sucks having to mix a WWDC presentation, sample code, and iffy docs into a semblance of usable architectural documentation.