We present our analysis of 254 Calculus I final exams from U.S. colleges and universities to identify features of assessment items that necessitate qualitatively distinct ways of understanding and reasoning. We explore salient features of exemplary tasks from our data set to reveal distinctions between exam items made apparent by our analytical framework. In doing so, we offer commentary on the often subtle features of assessment items that contribute to variation in the reliability of inferences an instructor is positioned to make about students' understanding based on their responses. We conclude by synthesizing basic principles exemplified by our results that mathematics instructors can apply to their assessment design.