One possible explanation is that the system’s optimization schedule is not well-suited for Monte Carlo training. This could be addressed by modifying the schedule to better balance the exploration-exploitation tradeoff, allowing the system to more effectively update its internal models and avoid plateaus in performance.