I'm still confused about the "repair" part of "race-and-repair". Could somebody explain whether/how this is different from eventual consistency in the limited c

I'm still confused about the "repair" part of "race-and-repair". Could somebody explain whether/how this is different from eventual consistency in the limited case of a database? Considering that you would (I guess) need to spend just as many cycles calculating the correct answer anyway, doesn't this make race-and-repair pointless for computations where correctness is required? And yes, I realize that a large part of the (cultural) problem is an overly high tendency to think correctness is required. So what kind of computations don't require correctness? The analogy to biological models is interesting, and bringing up cellular automatons (really agent-based models) gave me some ideas, but I'm still unsure where else something like race-and-repair would be useful.