Gravar-mail: Decoding the Formation of Reward Predictions across Learning