Ask, Don’t Judge: Binary Questions for Interpretable LLM Evaluation and Self-Improvement