This dissertation examines why military decision makers struggle to evaluate their policies and why they often stick to unsuccessful strategies for so long. The core argument is that strategic assessment involves genuine analytic challenges which contemporary scholarship typically does not take into account. Prominent theoretical frameworks predict that the longer decision makers go without achieving their objectives, the more pessimistic they should become about their ability to do so, and the more likely they should be to change course. This dissertation challenges those ideas and explains why we should often expect the very opposite. The theoretical crux of this argument is that standard models of learning and adaptation (along with many people’s basic intuitions) revolve around the assumption that decision makers are observing repeated processes, similar to the dynamics of slot machines and roulette wheels – but in war and other contexts, decision makers often confront cumulative processes that have very different dynamics, along with a different logic for how rational actors should form and revise their expectations. Empirically, this dissertation examines U.S. decision making in Iraq, Vietnam, and the American Indian Wars. These cases demonstrate how cumulative dynamics affect strategic assessment and how understanding these dynamics can shed light on prominent theoretical frameworks, ongoing policy debates, and salient historical experience.