We employ a novel experimental design to investigate how gender operates through distinct mechanisms in evaluating professional success versus failure. Using advanced artificial intelligence computer-generated imagery and motion capture techniques, we manipulate the perceived gender of tennis players while maintaining identical performances. Involving 3,755 representative U.S. Prolific participants, our experiments reveal how gender transforms between serving as a reference point in success evaluation and becoming a source of attribution bias in failure assessment. Study 1 examines physically demanding successful performances, showing that evaluators appropriately use gender as a reference point for calibrating expectations based on relevant group-level differences. Study 2 investigates basic execution errors where such differences are irrelevant, uncovering how gender becomes a mechanism for discrimination: identical mistakes receive significantly different evaluations based solely on perceived gender - errors are judged more harshly when perceived as made by women and more leniently when perceived as made by men. These findings advance our understanding of when gender appropriately informs versus inappropriately biases performance evaluations, extending regulatory focus theory's prevention-focus expectations and role congruity theory's status beliefs to explain this contextual transformation.