In model fitting platforms, the current (JMP18) row diagnostics options are Actual vs Predicted, Residual vs Predicted, and Residual by Row. The residuals are in the same units as the Actuals and Predicted.
I fit a lot of models to industrial process sensor data I can't be the only one that thinks about the residuals and tries to convert them to relative error %. i.e. I know my actual value has a relative error of x%. Is the prediction error greater or less than the measurement error of actual?
So, one improvement would be the plot that I have attached to this post: Parity plot + Residual in Engineering Units + Residual in % of Actual. A second improvement would be to allow the user to toggle between Predicted and Actual as the x-axis of the attached plot.
