This project draws together work on learning dynamic causal networks from simulation data with the application of domain knowledge to improve the model. In previous work we showed how causal networks capture domain expertise and can improve simulation modeling. We demonstrate this approach in this paper as it applies to a rudimentary reinforcement learning (RL) solution. Our thesis is that a person can understand the RL solution by means of the model causal structure derived from historical data. This is work-in-progress toward the larger goal of using the combination of domain knowledge applied to causal models to develop improved dynamic treatment policies.