I guess the question is whether we want universality over all limit-computable predictors, or only those derived from lower-semicomputable semimeasures. Since t

I guess the question is whether we want universality over all limit-computable predictors, or only those derived from lower-semicomputable semimeasures. Since the latter class contains its own universal element and also every computable predictor, maybe it's okay? But yeah, I can see how you can have a different takeaway. Sorry for being a bit harsh in the previous comment.