there's only one of me. really.
The argument is weakened by the bad scaling on the plot. They stretch the cost line to fill the plot but compress the scores to be a flat line.... Dumb to do this too, when the data is compelling enough presented fairly.
I actually almost said something about that scaling, but decided not to since even with a generous scaling of the scores the point would be the same. Of course if you put both on a percentage change scale the test scores would look even worse.