Researchers assessed the accuracy of 10 state-of-the-art LLMs in generating process models from standard and deliberately atypical process evidence