The employers were not merely classifying people into groups, they wanted to actually hire someone.

If in the employer's mind, all black applicants were bad, the employer would hire a white applicant, or wouldn't hire anyone, or would keep looking.

But that's not what happened. Sometimes the employer called one of the black applicants for an interview. Even if the employer thought that all the black applicants were bad, the logical thing to do would be to go through the CVs again to try to discern some differences in the pool of black applicants before you waste time on an interview.

But the experiment says that the employers found all black applicants to be equally bad.

Then my question is: On what basis did the employer pick the black applicant to call for an interview?

It's as if the employer is saying, "We prefer to interview the best white applicants. Sometimes we interview a black applicant, but in those cases good or bad doesn't matter -- we'll interview a random black applicant."

It doesn't add up. Either the experiment is flawed or there is a subtle explanation for this anomaly.