Flaw in Score Counting Algorithm

I have used Manhattan Tests in past. My personal experience was not satisfying. My average mocks score was 650 in Manhattan, but my score used to drop in Gmac original mocks. No doubt the questions in Manhattan both for quant and verbal are the best in the market, but the score calculating algorithm is what i have trouble with. I personally feel that even if you get consecutive wrongs in Manhattan your score would still not fall, on the other hand if you get wrongs in a row in Gmac mocks your score would drop. Manhattan mocks tend to give you an inflated score. So this is the only problem i have with Manhattan mocks.
I would say again the questions both in verbal and quant are the best out in the market.