When people do prediction, they also simulate system states given known priors and behavioral characteristics. The distinction you're making is in my opinion not valuable (or at least, you've not demonstrated it's value here).
Sure, and the point of simulation in the first place is often to understand how the system in question will behave, in advance. The distinction I meant to make is precisely that some processes may not be computable in this way; that they cannot be simulated any faster than the real thing, in other words. That is what is meant by computational irreducibility, AFAIK. Determining which systems this is true of is very valuable, imo
