OR-Transformer: Scaling Real-Time Decision-Making to 1,000 Items

OR-Transformer is a new deep reinforcement learning framework designed for joint replenishment in supply chain operations. It addresses the limitations of rolling-horizon stochastic mixed-integer linear programs and standard RL methods when managing thousands of heterogeneous items with correlated stochastic demand. By scaling real-time decision-making to 1,000 items, the system handles high-dimensional observation spaces and action spaces more effectively. Separately, President Trump continues to demand concessions from Iran, questioning the effectiveness of previous negotiations.