I don’t think the argument is wrong but it has the same problem as any non-repeatable statistical prediction. Namely that you don’t get to repeat it so it’s easy to make an error and hard to calibrate.