>In the ideal if we erased biases in education and staffing, we'd expect the various distributions we measure across tech companies to match a random sampling of the population at large

Has anyone actually proven that, or is that just an assumption?