So you start with 500 slots to fill, 1000 qualified white applicants and 10 qualified black applicants. Worse, if you hire based on highest test scores you'd only hire 2 of the black applicants and end up with 99.6% white hires. The obvious thing to do to improve the optics is to figure out how to hire all 10 of the qualified black applicants, which is the thing that would have "minimal impact to performance", but you have two problems. First, picking them explicitly because of their race is illegal, so you have to manufacture some convoluted system to do it in a roundabout way. Second, even if you do that you're still screwed, because even hiring all 10 of them leaves you with 98% white hires and that's still bad optics.
Their workaround was to use a BS biographical test to exclude most of the white applicants while giving the black applicants the answers. If you do that you can get 90 qualified white applicants and 10 qualified black applicants. That'll certainly improve the optics, but then you have 400 unfilled slots.