For instance suppose (totally randomly) than women tend to be better at reading comprehension tasks. Now if you knew nothing about someone and were told they were a woman this should make you increase your expectation of their reading comprehension. Now suppose also that women tend to test better than men, e.g., statistically they take tests more seriously and score higher relative to their inherent ability than men.

Now suppose you know only that someone got a 700 on the language section of the SAT. In this case further discovering that the individual is a woman should likely reduce your estimate of their reading comprehension since the effect of knowing that their SAT score is an overestimate will swamp the effect of knowing they are innately more likely to have a higher reading comprehension.

Now this point may apply just as well in the reverse. For instance if women are more intimidated by math exams than men (which some studies suggest is true) in fact finding out an applicant for math grad school is a woman should actually increase one’s expectation of their ability.

Ultimately I have no idea which way these conditionalization effects will fall but I do think we have good reason to believe these will overwhelm any innate factors. In short if we want to learn what effect knowledge of gender should rationally have on our judgment of someone we need to look to the interaction of gender and the information we are likely to get about them not the innate effects of gender.