I would be pretty amazed if the test didn't indicate bias between a complete A-A test (i.e. associate white people and white people with red words and green words), even leaving aside consistent scores as they discuss in the article.

Which means it's not reliably going to indicate anything in particular by itself. This is a challenge for a lot of these psychology tests and is part of the root crisis of reproducibility.