Simulated Telemetry Dataset for Privacy-Preserving Causal Digital Twins in Pharmaceutical Logistics

Global pharmaceutical supply chains suffer from physical-to-digital trust gaps, relying on reactive tracking that fundamentally fails to prevent cold-chain spoilage. While decentralized networks promise immutability, enterprise adoption remains hindered by the Oracle Problem, inadequate data privacy, and managerial algorithmic aversion. We engineer and evaluate a privacy-preserving causal digital twin architecture to execute autonomous logistics rerouting without exposing proprietary data. Our holistic framework integrates a fully homomorphic encrypted Graph Neural Network (GNN) with conformal prediction bounds to forecast disruptions causally. We bridge this predictive model with deterministic Web3 execution utilizing Zero-Knowledge Proofs (ZKPs) and account abstraction. The architecture is rigorously validated via 30-day stochastic multiverse simulations processing over 12.4 million telemetry logs, alongside managerial qualitative assessments. Simulations demonstrate our conformal-bounded GNN achieves a 74.00% topological prediction accuracy, with do-calculus mathematically proving that predicted structural shocks causally alter payload spoilage probability by 97.93%. Furthermore, integrating account abstraction enables instantaneous, autonomous rerouting execution, optimizing gas overheads below a 5% payload preservation threshold (sub-$0.50). However, qualitative evaluations reveal managerial algorithmic aversion increases during autonomous execution, heavily moderated by perceived ethical liability. We demonstrate that resolving the latency-privacy paradox requires concurrent risk attribution frameworks, providing a blueprint for trustless algorithmic capabilities in high-stakes logistics.

Note: This dataset contains 12.4 million rows of simulated IoT telemetry used to validate conformal-bounded Graph Neural Networks and DoWhy causal calculus.