And how do you determine who gets to be mentored in an internship?

Unless you take all candidates, you are back to reducing the pool to a more manageable size, presumably using classic interview techniques.

And unless the mentors are biased, this still doesn't change the fact that your hires as a population will be similar to the applicants.

> if the input pool is 98% men of course they are going to get 98% men out

Will they? What if women stay in their roles longer than men? Then, you would less often be replacing women.

If men more often apply to new jobs, that could explain why the candidate ratio is more skewed than the ratio of actual workers.

It's some of undeterminate without data on longevity because on the other hand we could hypothesize that even with equal gender representation initially maybe men would stay longer because women want to spend time raising children and that would skew things. Also, there might be some self filtering as you described in the last sentence where I could speculate that maybe the average man is used to being rejected a lot so they apply more to places that are hailed as hard to get into, though with the 30% figure female representation in CS touted elsewhere in this thread, it leaves the question where are these 30% women going to in CS?