That's essentially the plain-language interpretation of what the author of this article is pointing out - when you plot (actual score) against (difference between test score and actual score), you will always find a trend that underperformers overestimate and overperformers underestimate - for the exact reason you state.