tomorrow

very well said but you're missing my point. I wanted to emphasise any given LW user. Although particular LW users are very good at predicting, they do not appear to be the same ones who take the effort to vote, on aggregate

What are those other factors, that predict karma well enough that knowing whether what's said is right yields no further improvement in karma prediction on top of them?

In my model, karma is already overdetermined, and it's not a very good model, but similar factors that describe human behaviours as is cannon on LW go into that model. I may ellaborate in the future but like I said, probably not worth anyone's time and I'd rather not clarify on it myself than do another thing.