It's also the case that evaluating task performance is still a largely unsolved problem in many fields (e.g. programmer job performance). We don't have good obj

It's also the case that evaluating task performance is still a largely unsolved problem in many fields (e.g. programmer job performance). We don't have good objective metrics in many fields to evaluate "success" and in many cases the best we're able to do is the somewhat tautological "let the market decide" - what is good performance is just what is successful even though there are often many other factors and a large amount of luck and path dependence involved in that success.