Humans constantly construct intentions and act on them, but neither the content of intention nor the method of its construction are well-understood. We develop a series of tools that enable us to inspect intention and its construction even when our subjects are confronted with experimental disturbances. We then formed a predictive model of intent as a state trajectory constructed by a stochastic process that minimizes the cost of action while maximizing both reward and the rate of reward. This model appears to predict and explain the statistical distributions of both the pieces composing intent and and whole motions that we observe.