I acknowledge that effect is present, but it's simply not strong enough to dominate natural variation in ability (for which high school entrance exam scores are a proxy) to the point where the results are completely flat -- in fact, students who achieved the highest scores on the entrance exam were slightly lower in exit scores on average than those with low or medium entrance scores!

What this tells me is that some kind of correction is being applied to those y values (we already know this because of negatives numbers on a "percentile" axis) -- but what?

A lot turns on this. (I suppose I could read the underlying paper, but I'm lazy.)