Gravar-mail: Testing the Reward Prediction Error Hypothesis with an Axiomatic Model