Evaluating LLM-Based Process Explanations under Progressive Behavioral-Input Reduction