Judging AI-native work: why recognition authority should stay human