Supporting code and data for "The Gains from Uncertainty-Aware Ensemble Fusion in Imbalanced Fraud Detection Are the Weights, Not the Calculus"

Supporting code, results, and reproduction scripts for the article "The Gains from Uncertainty-Aware Ensemble Fusion in Imbalanced Fraud Detection Are the Weights, Not the Calculus" (H. A. El-Ghareeb). A controlled, leakage-safe evaluation of whether uncertainty-aware ensemble fusion improves imbalanced fraud detection; the finding is an honest map: the uncertainty calculi are interchangeable (no rule beats optimised weighted averaging by more than 0.001 AUC-PR), the gains belong to the weights and arise on simulated rather than real data, and uncertainty earns its keep on calibration rather than ranking. The archive contains the leakage-safe pipeline, all fusion rules and base learners, the statistical procedures, the aggregated per-configuration results, the figure- and table-generating scripts, and the scripts that retrieve the five public datasets - everything required to reproduce the study. This archive is the sole citable artefact; no third-party code repository is referenced. Code released under the MIT licence (see LICENSE).