Okay, pick some other metric – the lack of grade improvement starts to cut away at the possible claims for cognitive improvement, on the whole, assuming the science is good.
On the other hand, if they were sloppy, it could potentially argue for a multimodal distribution that when forced into being viewed as an average (or with other tools that assume a unimodal distribution) then the effect gets washed out. One possible case for this is the idea that ADD/ADHD is wildly over diagnosed and, thus the incorrectly diagnosed children might show no improvement while masking the improvement of those who actually have it.
Another anecdote but a very applicable one: a fellow nerd in high school had terrible grades, before and after starting Vyvanse (re-patented version of Adderall) around the end of sophomore year. However, he was able to focus in the biochemistry class and despite the fact that he still received bad grades for lack of caring, his medication helped him discover that he loved chemistry. A year later he was accepted to an Ivy League on a 3.0 unweighted GPA because he was able to throw a never before possessed focus at an internship at Caltech where he became a published scientist within a year.
Grade improvements (along with the possibility that over prescribing is skewing the data) are a terrible metric, and a better one would be interpersonal relationships and how they behave in the classroom, which is much harder to measure. If a child is diagnosed pretty late (late elementary or middle school) they might also have a LOT to really catch up on. Just because they can focus doesn't mean they are prepared to deal with the material given to them.
On the other hand, if they were sloppy, it could potentially argue for a multimodal distribution that when forced into being viewed as an average (or with other tools that assume a unimodal distribution) then the effect gets washed out. One possible case for this is the idea that ADD/ADHD is wildly over diagnosed and, thus the incorrectly diagnosed children might show no improvement while masking the improvement of those who actually have it.