There some good arguments against the writers' conclusions in the same comment section. From what gather, is he is saying that Frank is using the wrong data set, because the total casted in a county don't match the reported number. To that I say is so? All he had to do to get an acciarte prediction is to use the data set used im the algorithm, so if he gets a data set that gives him 99.9 % correlation then there is the money shot. Whoever input the data set into the machines used the same data set Frank used.