I am reminded now of the problems with research-level Q&A. There are questions which are answered in a relatively straightforward way, and these distort the truly difficult questions, which are those where the information is specialized, so that only a few people really know the answer. The reputation gain from answering easy questions always swamps the reputation gain from difficult research-level questions.
For reviews, the idea was to have two criteria, originality and correctness. Perhaps for questions, we should have two criteria too: "Researchiness" and "quality". The "researchiness" is the difficulty level, and the answers to difficult questions can come with exponentially more weight than the easy ones.
I would like to keep graduate level questions, but a question like "Why is G/H a manifold?" which is just a matter of shuffling definitions around, needs to be answered with a very small rep gain from both question and answer, as compared to questions about unsolved problem in fractional spin particles, which takes a heck of a lot more effort, and needs to give the questioner and answerer exponentially more reputation.
If you don't have "difficulty amplification", you will have a site dominated by low-level nonsense, rather than research questions, I think.
In this way, I think you can accomodate both kinds of questions, grad-level (with low researchiness) and professional level (high researchiness), and allow the community to determine researchiness by voting.
The mechanism is exactly the same as for reviews, and the scoring can be the same: exp((researchiness)^1/3 * quality), so that we don't punish good low-level questions, but we mark them as low-level.
Then professionals can choose read only the questions with positive researchiness score. It's an idea, but I am already worried from our first handful of questions that the site will not be professional level, and I don't want to delete low-level questions, simply to mark them as easier.