I mean nothing further is done to the targets for scoring – they are just left as-is. They are already 5 discrete equally-spaced values, so yes they are “ranked” in that way. Whereas the predictions the ties are broken (by “first” method). And then pearson corr is done on those two vectors, just as jrb said above and I’ve said a gazillion times. (True spearman correlation is actually never used in scoring, despite being referred to very often.) This has been discussed a million times around here, and the actual scoring code is posted. (And why are you always trying to tell me I’m wrong when I give you simple known facts about the tournament? I don’t make things up.)
You can even see in this thread New Target Nomi Release that I pointed out the 0.91 thing when we switched from the old target (where the max score was 0.98).