Because it's "standardized" testing, that's the whole point. If the nature of the test changes drastically yearly, how can you compare results from year to year?
It's "standardized" as in everyone gets the same test in a given year. Absolute results on something like a college admissions exam are pretty meaningless since getting into college is a zero-sum game for all but the worst schools. All that matters is doing better than X% of the population.
Seems to me like incentivizing deep learning over memorization outweighs losing the ability to compare absolute results from one year to the next.
