When a question is placed in the experimental pool, it isn’t assigned a difficulty level. It pops up at random on GMATs, whether the test-taker is getting a 350 or a 650. Eventually, a large population will have seen and attempted that question, and the testmaker will have some idea of its difficulty.
But that isn’t all they’re trying to do: they also want to determine whether the question is testing the same things that the rest of the GMAT is testing.
Here’s what I mean. Let’s say a problem appears as an experimental question on 1,000 administrations of the GMAT. It’s answered correctly 200 times, incorrectly 800 times. In other words, it’s about a 640-level question.
Ideally, the people getting that question right are scoring between 640 and 800, while the people getting it wrong are scoring between 200 and 640. It’s never going to work that way in practice, but the goal is to get close.
The point to take away from this is that, on average, experimental questions are of average difficulty. If you score a 700 on the GMAT, most of the questions you see will be difficult. In fact, most will be among the hardest 10% of questions in the database.