The GMATPrep software is as accurate a practice test can be.

A scaled score of 47 represents somewhere around 70th percentile right? So it looks ok to me.

Those official practice tests, like vineeshp said, are as accurate as you're going to find. And your theory that you were getting beaten up by hard questions is exactly right - to determine your score, the algorithm needs to pinball you between your upper threshold (you're probably not getting these right) and the maximum threshold that you just about always get right. So in order to score 700, theoretically you're going to see a bunch of 670-690 questions and get most of those right, but then see quite a few 710-730 questions and get most of those wrong. The scoring system is testing you almost by saying:

Slightly above your threshold: "Okay, hotshot...you got that 680 question right but let's see if you can do 720"

...you get that wrong, so they give you

Slightly below your maximum level: "But surely you can handle 690, right?"

Obviously, there will be some false positive (you guessed right, even though you didn't really know it) and false negative (you should have had it but made a silly arithmetic mistake and ended up estimating wrong to save time) results, so they'll take all 37 questions to work through this process. But ultimately what you need to know for this week is:

1) It will feel like a difficult test regardless of how you're doing. They have to test your upper limit repeatedly to determine where you "max out", so you'll feel pressured quite often, and that's a good thing.

2) You can afford to miss questions. Everyone misses quite a few - if you score 730, you end up missing a lot of 750-760 questions...that's just how it goes. So don't beat yourself up if you need to guess once or twice...it happens.

3) If you're taking the test properly, you don't have time/energy/expertise to be able to predict the difficulty levels of questions. Many "hard" questions look pretty easy...the difficulty is kind of sneaky. Some easier questions look hard, but are actually pretty manageable if you see the opportunity to get away with a number property/estimate/backsolve. Your job is to get questions right - not to determine what your score looks like at that point. I've seen multiple cases of someone psyching herself out by looking at a question and saying "oh, no...it's too easy...I must be doing poorly" and then unraveling from there. There's no point in trying to determine how you're doing - it really can't help you and I know for a fact that it's severely hurt many students. The test will score you properly - your job is just to answer the questions to the best of your ability.

If you want to read up more on the scoring system, this article should help:

https://www.veritasprep.com/blog/2010/08 ... algorithm/

Thanks, that's a very helpful explanation.

I figured that the two official practice tests would be basically the same as a the actual exam, but I was surprised that my experience was so different from some of my practice exams through unofficial test prep sources. On one unofficial practice test, I only missed one question and the scaled score came out a 60--ridiculous, right? They just don't have enough difficult questions on that test.

I can complete most of the questions, but on this official test I was taking too long to answer early on, which really hurt me towards the end. I guess it becomes a balancing act of whether to rely on process of elimination to get my odds down to 50/50 or to spend an extra minute or two on a problem and narrow it down to the correct response. I think I tend to take longer on the data sufficiency problems because it can sometimes be 3-questions-in-1.