Maria assumed talent is evenly distributed amongst the population. But it’s not.
If the candidate developer cohort includes, say 10 women and 90 men, then talent will be normally distributed (bell curve) over the population of women, and normally distributed over the population of men.
If, for social engineering purposes, you dictate that all 10 women must be hired, and then you’ll fill the remainder of roles from the male applicant pool, you have made a bad selection. Why? Because you hired the entire talent spectrum (untalented-talented) from the female pool.
You should have used a single talent-based cutoff and used this to select from both gender pools equally - even if that meant taking in a smaller number of women than met the criteria for your social engineering experiment.