Score one set of predictions four ways. Add a few outliers, a systematic bias or errors that grow with the value, and watch MAE, RMSE, R squared and MAPE react differently. See why RMSE over MAE equals 1.25 for Gaussian noise and rises toward the square root of n when one miss dominates, and why MAPE is capped for under-prediction but not over-prediction. A Blender-built plotting easel with Gaussian-ratio, exact-inequality, correlation and constant-ratio checks built in.