Authors - Vishwa Kumaresh Abstract - A local supplier delay or demand shock in multi-echelon supply chains can make upstream orders volatile long before the full costs appear in planning dashboards. In this study, we propose an interpretable warning-to-action layer for supply-chain digital twins. This layer sits above the replenishment controller: it estimates disruption-regime risk from rolling demand, inventory, backlog, order, and lead-time telemetry, then maps that risk to bounded changes in responsiveness, safety stock, and order caps. We calibrate a gradient-boosted stump classifier that combines standard warning indicators, cross-echelon imbalance measures, and nonlinear stress descriptors. A small mode table converts the resulting probability into five auditable replenishment modes. This method is tested on twelve disruption scenarios grouped into six mechanism classes, using ten baselines and an untouched lockbox of 576 observations. The proposed policy reduces aggregate system expenditure by 15.2% and cross-echelon volatility (bullwhip) by 44.5%, relative to a linear guard that uses the same broad action family. The largest gains occur in lead-time disruptions and backlog cascades. Compound shocks demonstrate marginal performance gains, as existing linear guards effectively capture these dynamics within standard monitoring frameworks.
Open Zoom