Given the authors second point was math, he or she could have at least done it honestly.
Lets go with the 25,000 undervotes. 70% to Nelson is indeed 17,500 but ignores the 7,500 Scott would have gotten. Precinct level analysis could possibly yield a more precise answer (not going there in terms of which side of the electorate would be more apt to screw up) but in all likelihood Nelson would still have lost.
Good point.