And then would be very interesting to feed a Bayesian filter with this feedback, and see if that filter can pick up future "snarky" comments automatically.
If that works, then requiring a proof of work (a simple-ish logical puzzle) attached to a snarky comment could bring a scalable barrier.
Sometimes it can be difficult to reply to a comment without the possibility of being perceived as being snarky. For example a comment thread that I was involved in about jQuery recently:
Fortunately the person I replied to took the comment in the sense it was intended, but I also know that people can be really defensive in this area and can perceive a comment as snarky even though it wasn't meant to be.
I read it and do not find anything snarky in it. But the way you mentioned it made me wonder: there are "intentions" and there are "results" - are there a lot of the comments that are intentionally snarky ? Or, the snarkiness is just a way a person would communicate ? And if that, is it worth hinting to that person that their written communication conveys not what they intended ?
If that works, then requiring a proof of work (a simple-ish logical puzzle) attached to a snarky comment could bring a scalable barrier.