CAUSE-RAG reproducibility materials for causally audited time-series forecasting explanations

This record contains the reproducibility materials for the manuscript “CAUSE-RAG: Causally Audited Universal Structural Explanation through Retrieval-Augmented Generation for Time-Series Forecasting,” submitted to Knowledge-Based Systems. The archive provides the data, model, evaluation, figure, and manuscript-supporting materials required to reproduce and inspect the CAUSE-RAG experimental workflow. CAUSE-RAG is a causally audited knowledge-based retrieval-augmented generation framework for explainable and grounded time-series forecasting. It combines a hierarchical causal registry, typed numerical grounding, counterfactual reasoning, post-hoc explanation fusion, and release-time auditing to restrict generated explanations to registry-grounded numerical claims. The uploaded bundle includes hydrological forecasting datasets, trained model checkpoints, scaler objects, evaluation outputs, final manuscript figures, supplementary figures, metadata files, and repository documentation. The associated GitHub repository hosts the structured release package and will be updated with the full experiment code. These materials support the reported evaluation across six Northumberland catchments, three temporally disjoint zero-shot windows, eight post-hoc explainers, ten retrieval-augmented generation strategies, sixty RAG baseline configurations, and six language-model backends.