In the context of tests being biased, I'll just note that your idea amounts to (unbiasedly) testing the test designers.

Quis custodiet ipsos custodes?