A reinforcement learning method that groups multi-turn decisions between experiment results into windows and assigns shared rewards based on outcomes.