.
This is Classical Signal Detection — Predictive Analytics — Machine Learning .... type of problem. It is the standard Precision vs. Recall accuracy in any learning model ..... Perfection is not possible, nor is it desired, because every system process is inherently Stochastic in nature, and we do our best to work with the noise that is ever present in everything .... including Malformed ambiguous SAT questions.
700 is about the “Hinge-Point” in the Precision/Recall curve for both SAT math and verbal sections. Half of the questions that were missed getting to this plateau hinge point in the curve are mostly flip-a-coin questions that were not well written, and can go more than 2 possible ways.
With this in mind, a test taker should practice and know before hand what their Benchmark or basis is before taking even the 1st Official test. Practice tests and Kahn revealed that our daughter needed no additional practice on the SAT verbal. She was already at that plateau of 700. I worked with her on the Math section for about 2 months before she took her 1st test, identifying mostly strategy issues and the way she approached certain problems ... like quickly Doodle-Graphing linear inequalities as opposed to more time consuming substitution or factor elimination techniques .... Adding more Tools to her Math problem solving toolbox helped her tremendously.
Conclusion .... 700+ Math/Verbal scores are very good scores .... and if they are on gotten on the same day in the same sitting (no superscore), it is that much better, because it really is a Marathon.
Anything at 750 and above is really wasted effort as far as the Statistics go, because you start working in the Noise region of the Stochastic problem.