A judicial experiment replicates a study on 61 U.S. federal judges using GPT-5 as the decision-maker in a hypothetical automobile accident case. The experiment manipulates three variables: the applicable doctrine as a rule or standard, and the portrayal of the plaintiff or defendant.