Run a forecasting policy against live demand that carries trend, season and the occasional anomaly. Capture the structure, keep smoothing moderate, and don’t let one spike whip your forecast. Pick a method, tune the sliders, then press Run (or Step for one round at a time) and beat Par over 20 rounds.
Demand carries a trend and a repeating seasonal shape, plus random noise and rare one-off anomalies. A good forecast tracks the structure without over-reacting to a single strange period.
Each round the engine draws real demand and your policy forecasts it from the smoothed level and trend. Error (MAPE) and service level are scored against holding cost from your safety stock.
Model trend and season, keep α moderate so anomalies don’t whip the level, and hold only as much safety stock as service needs.
