In a test of 150 impasse samples, the baseline tutor gave direct answers in 50.7% of its responses, the no-direct-answer tutor posed follow-up questions every time, and the Socratic tutor gave diverse responses based on context.