"X correlates with race" and "X correlates with culture, which correlates with race" are the same thing. If neither X nor culture are related to what you're try

"X correlates with race" and "X correlates with culture, which correlates with race" are the same thing. If neither X nor culture are related to what you're trying to measure, it's still just racial bias. We've been using "ice hockey" as an example, right? Given that it's not obviously correlated with fitness for a job at Amazon, it's incumbent on you to show that it actually is. Until then, we should assume that it's at best a false positive from unrelated data (like the ruler in the skin cancer example) and at worst a proxy for race and gender.