So you seem to be arguing that we should use a model, but a less accurate model. I think if you're going to use a model, you should use the best one(s) you can. You say prior performance is inherently unfair. I don't understand why. No college football team turns over 100 percent of their players. A Bayesian model starting with the prior that everyone is equal would likely be very poor. It doesn't bother me that there is some weight given to the prior given the limited sample size. There's probably room for reasonable disagreement on how much weight to give, but my arbiter would be predictive accuracy. But if it violates your sense of justice (and you're not alone) then we'll just have to disagree on that point.