Brain-Machine Interfaces (BMI) actively learn to control a prosthetic device using a relatively small number of neurons. Once the task is successfully learned, what information can be extracted from the BMI control mapping? Most BMIs reconstruct trajectories and ascertaining neural contributions is possible via sensitivity analysis of the input-output relationship. Value-based (a subset of goal-based) BMIs judge possible actions to find a best action at each time step and the sequence of selected actions will form a trajectory. Here, we expand the sensitivity analysis of trajectory-based BMI such that it applies to value-based BMI. Our finding that only a subset of recorded neurons contributes most to prosthetic control agrees with prior BMI research. Additionally, we find some specialization in neurons, i.e. certain neurons contribute to a subset of actions while other neurons contribute to different actions. Finally, we discuss implications of this metric and areas for future improvement.