Search

Reproducibility has emerged as a critical bottleneck in data-driven materials engineering, particularly as graph neural networks (GNNs) and associated uncertainty quantification (UQ) methods are increasingly embedded in high-stakes discovery pipelines. While advances in conformal prediction, Bayesian inference, and ensemble techniques have improved predictive reliability, their practical deployment remains constrained by fragmented workflows, opaque data provenance, and inconsistent reporting standards. This review reframes uncertainty-aware materials modeling through the lens of reproducible workflows, tracing the pipeline from dataset construction and curation to model training, calibration, and deployment. Drawing on peer-reviewed studies, we integrate methodological advances in UQ with emerging practices in data governance, experiment tracking, and model documentation. The analysis reveals that uncertainty estimates are only as trustworthy as the workflows that generate them: biases in dataset composition, undocumented preprocessing steps, and inconsistent calibration protocols systematically undermine reliability, even when state-of-the-art UQ methods are applied. We argue that reproducibility must be treated as a first-class design constraint, requiring standardized data provenance tracking, version-controlled training pipelines, and model cards that explicitly document uncertainty behavior, calibration performance, and failure modes. By linking UQ theory with reproducible systems design, this work establishes a framework for trustworthy materials graph learning in which uncertainty is not merely computed but auditable, interpretable, and transferable across applications.

The computational and data-driven paradigm has fundamentally reshaped materials engineering, enabling the navigation of vast chemical and structural spaces through machine learning, high-throughput computation, and autonomous workflows. Yet this transformation has also exposed a critical conceptual gap: the absence of explicit, structured boundaries that govern the division of cognitive labor between human experts and artificial intelligence systems. Without such boundaries, AI contributions risk overstepping domains requiring physical intuition, ethical judgment, and contextual synthesis, while human oversight may inadvertently constrain the scale and speed that define modern discovery pipelines. This manuscript introduces the Epistemic Jurisdiction Framework (EJF), an original systems-level model that delineates jurisdiction layers, interfaces, and feedback mechanisms tailored to the materials discovery ecosystem. The EJF maps the flow from raw data to validated discovery through distinct zones of human primacy, AI autonomy, and negotiated hybrid spaces, emphasizing representation–inference interactions and computational steering logics. Grounded in the recent literature on machine learning for materials, explainable systems, and data-driven infrastructures, the framework offers a conceptual scaffold for infrastructure-level design rather than performance optimization. Its implications extend to the construction of more robust, interpretable, and sustainable computational ecosystems in which human and AI capabilities are aligned rather than blurred. The EJF thereby provides a foundation for next-generation materials engineering platforms that preserve epistemic integrity while fully exploiting computational scale.