Replication Data for "Stable Predictions, Fragile Coordination: Diagnosing Interpretation Risk in AI Decision Support Systems"

This dataset contains the analytical data and replication code for the DIV-DRO framework study on interpretation stability in AI-assisted decision support systems, applied to 544 forest fire incidents in Taiwan (FANCA, 2024). Includes cleaned event data, 20-seed split experiment results, pairwise Coverage Loss and Narrative Shift metrics (190 pairs), and distributionally robust optimization outputs.