Or Google can reduce the overrepresented candidate pool to match that of underrepresented by randomly select the appropriate number of candidates from the overr

Or Google can reduce the overrepresented candidate pool to match that of underrepresented by randomly select the appropriate number of candidates from the overrepresented (i.e. Randomly reject, lol). This should have no effect on te bar each candidate is expected to pass, underrepresented or not. Interestingly this hypothetical scenario would probably increase the false negative rate for overrepresented applicants due to random rejection, which effective lowers false negative rate for underrepresented applicants relative to the overrepresented.