[talltimtom]

Ok let’s take your simplified model assuming women and men are equally qualified but woman are less frequent due only to their choice, the top 10 are 7 men and 3 woman. Duolingo are looking for 2 new hires and hire 1 woman and 1 man. No problem.

Remember duolingo don’t hire 100% of the available workforce so they can stay 50/50 just fine, they are just snapping up the most qualified woman and making it harder for everyone else to follow suit.

[throwawaygender]

For nicer numbers, assume the bar is top 10% and you have a budget for 10 new hires.

Looking at 1000 people, there are 70 men and 30 women who are above the bar, from which you can pick 5 men and 5 women.

But these 10 are still not the top 10 within the top 100! This is because the distribution of the top N never changes, assuming men and women are distributed identically. It is still 7 men to 3 women. The 10 you've chosen are not the absolutely best top 10 you could've chosen without considering gender.

You can double the size of the pipeline (e.g., there are now 200 people above your bar) and the argument will not change. The minimum ranking of their incoming class is still lower than just picking the absolute top 10 without considering gender.

I accept that this tradeoff is worth it (and above the bar, the actual performance difference between ranks is negligible). But I hate that people will not agree that is a tradeoff.