Decoupling Disruptive Dynamics: A Foundation Time-Series Paradigm for Quantifying Multi-Tiered Supply Chain Resilience

Abstract

Quantifying supply chain resilience under volatile global markets presents a persistent epistemological challenge, primarily due to the non-linear convergence of disruptive shocks and subsequent recovery trajectories. This study develops a robust, data-driven measurement framework by integrating Long Short-Term Memory networks into a Hybrid Foundation time-series architecture, constructing a complete Dynamic Resilience Index that encapsulates both localized shock absorption and macroscopic elastic recovery capabilities. Utilizing a multi-variate operational dataset derived from 15 manufacturing enterprises across a four-year horizon (2020–2024), the proposed empirical configuration captures subtle variations in core performance indicators, including order fulfillment rates, inventory turnover dynamics, and multi-tier transportation latency. The empirical results demonstrate remarkable predictive performance, yielding a mean squared error (MSE) of 0.0019 and a mean absolute percentage error (MAPE) of 4.32% under baseline conditions. Rather than presenting an idealized operational trajectory, stress testing via noise injection and missing data reveals complex model behaviors, exposing latent biases related to systemic parameter friction and pre-trained structural generalizability. While alternative interpretations regarding endogenous structural shifts remain plausible, this framework offers an adaptable tool for structural monitoring and target interventions, demonstrating that hybrid systems significantly enhance operational predictability despite inherent architectural boundaries during black-swan events.