Indeed, often acquired directly, but most often supported through proxies, which often have multiple partial correlations, such as, median house price in the applicant's area, crime rate in the applicant's area, applicant's job title, etc.
That's a good point - maybe this is just the de-cloaking of what already goes on based on a really specific example. Will we generally accept it though, when it's laid out so plainly?
