In theory a test is a random sampling of knowledge or skills that are representational of a broader base of skills and knowledge. If you are trained only on the mode of the test, the test is only reprentational of itself and the narrow range of skills it's squeezed into a multiple choice format. With no test prep, then various touchpoints within become representative of a broader base of knowledge.
Imagining a fairly linear set of knowledge in math with 100 concepts numbered 1 to 100. If only the even numbered concepts are testable with a multiple choice test, and you had no prep, then getting a question right on each of the even numbered concepts is a fair representation of your knowledge of the odd questions, too. If you only prep, then getting all the even numbered concepts right is Not representative of your knwoledge of the odd numbered ones.
If you study geometry before someone else, then the test is still fairly measuring your knowledge if that is the intent of the test. If the intent of the test is as an IQ proxy, then it could be a poor proxy if the only barrier to learning geometry early is a time or resource or school constraint.
In most normalizing studies I've read they remove people who have prepped from the study. Meaning in reality they aren't properly normalized. If the only intent of a test is to predict successful graduation with a four year degree, then prepping seems completely appropriate as any advantages they convey are likely representational of similar advantages a student would have while in college whether resources or heavily involved parents. But they should include those in their normalization studies.
