Oh, and one other thing, at a meta level: The entire construction of the author's argument is to pick a statistical measure with a wide standard deviation, instead of a better one that would use more data to get a low standard deviation. Then he finds that one real-life outcome is within that wide standard deviation.
Not only that, he defines it via the exponentially increasing ELO rating distribution, instead of rank order, and that would have the effect of piling up a bunch of lower ranked players together within one standard deviation of the mean. This means it would be virtually impossible to get a result 1 standard deviation out, let alone 2.
Not only that, he defines it via the exponentially increasing ELO rating distribution, instead of rank order, and that would have the effect of piling up a bunch of lower ranked players together within one standard deviation of the mean. This means it would be virtually impossible to get a result 1 standard deviation out, let alone 2.