Why wouldn't your knowledge of how often you misread problems factor into your estimate of what the probability of your answer to the problem being correct is?

Why wouldn't your knowledge of how often you misread problems factor into your estimate of what the probability of your answer to the problem being correct is? The scheme is rating how well you answer test questions, not how well you know the answers to the problems. As you point out, there's subtle differences. But no grading scheme on earth could address those subtle differences, so it's not exactly a failing if his scheme doesn't.