How might we balance assistance and penalties to intelligent tutors and educational games that increase learning and interest? We created two versions of an educational game for learning policy argumentation called Policy World. The game (only) version provided minimal feedback and penalized students for errors whereas the game+tutor version provided additional step-level teaching feedback and immediate error correction. A total of 105 university students played either the game or game+tutor version of Policy World in a randomized, controlled, two-group, between-subjects experiment, during which we measured students' problem-solving abilities, interest in the game, self-reported competence, and pre- and posttest performance. The game+tutor version increased learning of policy analysis skills and self-reported competence. A path analysis supported the claim that greater assistance helped students to learn analysis better, which increased their feelings of competence, which increased their interest in the game. Log data of student behavior showed that debate performance improved only for students who had sufficiently mastered analysis. This study shows that we can design interesting and effective games to teach policy argumentation and how increasing tutoring and reducing penalties in educational games can increase learning without sacrificing interest.