I think your sample size is too small to predict with any accuracy. You need to test about 1000 tires before you can get any assurance that your results have any validity. As I recall, a sample size of 1,000 would mean your results would be accurate to within about 3% 19 times out of 20. We need a better way of doing flat prediction. We know that if you ride on any tire long enough, there is a 100% probability that you will puncture it. I would propose n km before puncturing would be a more meaningful starting point. The idea would be that you'd have a 95% probability that the tire would go n km before going flat. That last 5% accounts for completely random behavior. But the closer you get to km n, themore likely you are to puncture! So your Tire A would have a rating of 333 miles (533 km) before puncturing. Tire B would be 2000 miles (3200 km) before puncturing. Sort of a mean time before failure rating.

Luis