A judgment call can have more than one defensible answer: evidence does not dictate one correct option. Anthropic’s October 27, 2026 report tests influence by assigning the AI adviser’s favored option by chance, without telling analysts. Chance assignment helps avoid systematically giving opposing preferences to different kinds of people. Compare the shares choosing option A when the AI favors A versus B. A percentage expresses a share per 100; 60% versus 20% differs by 40 percentage points. This estimates average influence of the adviser’s direction, not the effect of receiving AI. It does not reveal everyone’s starting view or make an influenced choice a factual error.