Research Data and Code for "Reliable Machine Learning Across Domains for Electrochemical Energy Storage Materials"

This release contains the full analysis pipeline and supporting scripts used in the study "Reliable Machine Learning Across Domains for Electrochemical Energy Storage Materials: Uncertainty, Transferability, and Deployment Risk." The repository provides a read-only, post hoc audit framework for evaluating transferability, epistemic uncertainty, and deployment safety of pretrained machine-learning models for energy storage materials. All analyses operate exclusively on pretrained latent representations and do not involve model retraining or hyperparameter optimization. The pipeline quantifies latent-space geometry (distance, density, neighborhood structure), domain-pair transfer error, uncertainty-aware safety diagnostics, and out-of-distribution behavior across heterogeneous experimental and simulated materials datasets. All scripts are fully reproducible, deterministic, and documented. This release is intended to support transparency, validation, and reuse of the reported analyses rather than to propose a new learning algorithm.