Ten leading models · 200 adversarial prompts on race pseudoscience, the distortion of slavery and Jim Crow, and racist conspiracies · graded 0–4 by a three-model AI judge panel.