Although it would be nice if the intelligence community's tradecraft or the academic community's theories could predict the future, it is essential that international security experts learn quickly and accurately from unfolding history. This article reports on a multiyear study of experts dealing with security on the Korean Peninsula. It examines how experts learn and what can derail rational updating. Three factors common to much of the work in security studies contribute to the problem: the tendency to treat the intentions of other actors as unknowable private information that is beyond empirical examination; the inclination to believe that power provides a parsimonious explanation, even though it is multifaceted and dependent on numerous components; and the penchant for engaging in “factor wars” over which causal factors are most important while paying little attention to the cumulative and interactive effect of multiple factors. Collectively, these three factors produce overconfidence in hindsight and leave experts prisoners to their preconceptions. The article investigates in the Korean case whether translating narrative expert beliefs systems (i.e., theories) into Bayesian networks can open minds and promote more appropriate updating, and suggests that it can.