R squared 

I discuss the perhaps heretical way I like to use an R-squared style measure: as a function of the presented model performance alone applied out of sample, independent of details of the fitting procedure.