How did you choose the process noise covariance in your `grad` function? It doesn't seem like a single process noise covariance structure should be globally app

How did you choose the process noise covariance in your `grad` function? It doesn't seem like a single process noise covariance structure should be globally applicable across all possible functions.