I hear your pain, ddd!

It can be difficult to compare performance between different test makers because as you said, the algorithm works differently with each. If you want to gauge your improvement, it would be better to compare scores within the same test company: example: compare your scores only amongst your other PR scores.

If you want to get a sense of what your actual performance might be like, focus on GMATPrep.

Good luck! Keep focusing on your weaknesses!

Also, don't focus on percentage correct. On the real test, the vast majority of test-takers, regardless of score, get about half of the questions right and about half of the questions wrong. I haven't studied the PR algorithm but from stats you and others have posted, it's clear that this test doesn't mimic the real algorithm.

Think of your score as the difficulty level at which you can get half of the questions right - it's not the level at which you get all of the questions right. (Think about it: if you and I both get all 500-level questions right, does that mean we are 500-level test takers? Maybe - but maybe we can also get all 550-level questions right, too. So the CAT instead uses the 50/50 concept: the level at which you can get about half the questions right is "your" level.