The reliability of an AI research agent should not be judged only by its final results. We also need to evaluate how it handles evidence, makes decisions, and reports uncertainty. Two recent studies ...