How do we ensure AI doesn’t miss signals that you’d see in a practice?