A common narrative in physics education research is that students taught in lecture-based classes learn less than those taught with activity-based reformed methods. We show this narrative is simplistic and misses important dynamics of student learning. In particular, we find students of both methods show equal short-term learning gains on a conceptual question dealing with electric potential. For traditionally taught students, this learning rapidly decays on a time scale of weeks, vanishing by the time of the typical end-of-term post-test. For students in reform-based classes, however, the knowledge is retained and may even be enhanced by subsequent instruction. This difference explains the many previous pre- and post-test studies that have found minimal learning gains in lecture-based courses. Our findings suggest a more nuanced model of student learning, one that is sensitive to time-dependent effects such as forgetting and interference. In addition, the findings suggest that lecture-based courses, by incorporating aspects designed to reinforce student understanding of previously covered topics, might approach the long-term learning found in research-based pedagogies.