That sounds an awful lot like updating our prior beliefs based on data a la Bayesian modeling. So it's actually not a terrible way to go about updating noisy estimates.