We examine learning, model misspecification, and robust policy responses to misspecification in a quasi-real-time environment. The laboratory for the analysis is the Sargent [The Conquest of American Inflation, Princeton Univ. Press, Princeton, NJ, 1999] explanation for the origins of inflation in the 1970s and the subsequent disinflation. Three robust policy rules are derived that differ according to the extent that misspecification is taken as a parametric phenomenon. These responses to drifting estimated parameters and apparent misspecification are compared to the certainty-equivalent case studied by Sargent. We find gains from utilizing robust approaches to monetary policy design, but only when the approach to robustness is carefully tailored to the problem at hand. In the least parametric approach, the medicine of robust control turns out to be too potent for the disease of misspecification. In the most parametric approach, the response to misspecification is too weak and too misdirected to be of help. But when the robust approach to policy is narrowly directed in the correct location, it can avoid Nash inflation and improve social welfare. It follows that agnosticism regarding the sources of misspecification has its pitfalls. We also find that Sargent's story for the rise of inflation of the 1970s and its subsequent decline in the 1980s is robust to most ways of relaxing a strong assumption in the original work.