Cool. Thank you for the well thought out response.
Initially the potential problem I see with this is that it's completely arbitrary. The 'damage done' and 'good it brings' scales are totally subjective, and would probably even vary day-to-day depending on the decision-makers' moods. I think there is also the potential that over the long term the goalposts could shift, to the point where just generally being a prick is 'doing more harm than good' in the decisoon-maker's eyes.
That said, it seems judgement calls have to be made. And by extension, poor judgement calls have to be tolerated to some extent by the userbase. I believe this has already happened, as the admins are not perfect.
This is an incomplete answer to your proposition. I'll have to think it over a bit more.
Algorithms should be used to make all decisions pertaining to user restrictions, just like how they assign the 'suspicious behaviour' tag. If you push the suspicious behaviour hard enough the algorithm throws you in s/jail. Algorithms have no moods and can't do anything arbitrarily.
(post is archived)