This paper links the term structure to perceptions of monetary policy. Long-horizon forecasts of short rates needed in empirical term structure models are heavily influenced by the endpoints, or limiting conditional forecasts, of the short rate process. Mean-reversion or unit roots are commonly assumed, but do not provide realistic yield predictions. Failures occur because neither accounts for historical shifts in market perceptions of the policy target for inflation. This paper links endpoint shifts to a learning model where agents must detect shifts in long-term policy goals. With shifting-endpoint short rate processes, models better explain yield fluctuations.