A student gets an unusually low score, studies hard, then scores closer to their usual level. Did the study session cause all the improvement? Not necessarily.
Measurements contain variation. When a result is extreme partly because of random noise, the next measurement is likely to include less of that same extreme noise. It therefore tends to sit nearer the person’s typical value. Statisticians call this regression to the mean; it is a tendency, not a promise that every next result will be average.
The same pattern can make a remedy appear effective if people are selected for help right after an unusually bad measurement. Some would improve on a repeat test even without the remedy. A comparison group and repeated measurements help separate that natural settling from a real treatment effect.