Fair; StackExchange might not be the best model. I'd say public reviews of papers, but not crowdsourced ratings of reviews - rank reviews by researcher reputati

Fair; StackExchange might not be the best model. I'd say public reviews of papers, but not crowdsourced ratings of reviews - rank reviews by researcher reputation instead, weighting that reputation by reputation in a given subfield, and maybe with a weighting boost if the paper in question cites the researcher's work. Anecdotally, I've got a paper draft on ArXiv right now, which I sent to both the authors of the paper I based it on (who I got some really useful feedback from), and to a mailing list full of developers working on similar problems (who argued with me about typos my abstract didn't have) - this suggests that the people you cite might have better input into your paper than the general public.