A test setup that hides key identities and test details from both sides.
It is like a school contest with covered name tags. The judges cannot see who made each robot. The teams cannot peek at the course.
This makes scores less about fame or leaked questions. It is used for AI tests and preference reviews.
Evaluation harness
An evaluation harness can hide names, give tests, and record scores.
Third-party AI evaluation
Double-blind testing helps third-party results feel less biased.
Benchmark contamination
Hidden test questions make it harder to train only for the test.