For testing, I think what works best is doing a strict on-off regime for successive tankfuls, and then analyzing the resulting distributions for significance of difference. The beauty of this method is that the distributions tell you the reproducibility of your mileage, so you know whether the...