This paper provides the first evidence that value-added education accountability schemes induce dynamic distortions. Extending earlier dynamic moral hazard models, I propose a new test for ratchet effects, showing that classroom inputs are distorted less when schools face a shorter horizon over which they can influence student performance. I then exploit grade span variation using rich educational data to credibly identify the extent of dynamic gaming and find compelling evidence of ratchet effects based on a triple-differences approach. Further analysis indicates that these effects are driven primarily by effort distortions, with teacher reallocations playing a secondary role.