I've tested many of the most-cited websites in citoid, and results are generally positive but still have a lot of issues. (My very limited testing has led to nine open bugs; see also ) Given the vast different # of sites that could be fed into citoid, I suspect that there will be a long tail of sites that have bugs.
Given that, I'd suggest making it easy for power users (maybe autoconfirmed?) to submit bug reports about sites that don't work well - it could automatically pull the URL and citoid-generated cite, and give people a chance to explain which fields they thought were bad/incorrect/incomplete/etc.