I was sort of with the author until this point. He misunderstands the goal that the memo stated was at the heart of the issue, and thus his model is too simple to be accurate. The missing element was Google's false negative rate.
This isn't a case of selecting (for purposes of analogy) all gems above some certain quality, it is the case of selecting the best gems on an assembly line guaranteed to reject 90% of gems. If we reject 80% of yellow gems and 95% of green gems due to the fact that we have more green gems, we have created a lower standard of entry for the yellow gems, even if our gems are all top quality.
His analogy only works under the assumption that google has a reasonable applicant to hire ratio, but we know this to not be the case. Google isn't taking average (by Google metrics) Men or Women, and that's why the Women are very good.
Placing the male/female ratio as an input to your hiring process discriminates candidate hires based on sex, which is illegal in the United States.
Please someone correct me if I'm missing the point.