Anchored FLoE: A business-guardrailed ensemble framework of foundation and local-trained models for demand forecasting

Accurate demand forecasting is vital for retail supply chain efficiency, yet a persistent trust-capacity gap limits industrial production to low-capacity interpretable models that fail to capture complex market dynamics. We propose Anchored FLoE, a dual-model framework that bridges this gap by fusing high-capacity deep learning with rigorous business guardrails. The framework integrates: (1) FLoE, an ensemble merging a frozen Time Series Foundation Model (Chronos-2) for global generalization with a Local Specialist (TFT) for domain-specific patterns; and (2) a model-agnostic Anchor Layer that enforces economic monotonicity and business-logic consistency through discount-segmented confidence bands. Validated across 524 product-market segments, Anchored FLoE achieves an 11-percentage-point improvement in sales-weighted MAPE over production baselines, translating to a multi-million dollar impact on free cash flow.