Reproducibility Package for "Aggregate–Subgroup Calibration Discrepancy in Online Quantile Feedback: An Exact Error-Cancellation Decomposition and Empirical Analysis of Asymmetric Decision Loss"

This record contains the supplementary materials and reproducibility archive accompanying the manuscript “Aggregate–Subgroup Calibration Discrepancy in Online Quantile Feedback: An Exact Error-Cancellation Decomposition and Empirical Analysis of Asymmetric Decision Loss.” The study examines disagreement between aggregate quantile calibration, subgroup calibration, and asymmetric decision loss using Capital Bikeshare, Seoul Bike Sharing Demand, and Appliances Energy Prediction data. Evaluation covers four chronological origins and three target quantiles per dataset, yielding 36 dataset–origin–target settings. The package contains detailed methods and derivations, experimental results, saved predictions, analysis scripts, figures, reproduction instructions, and provenance and checksum records. The cancellation analysis is a post hoc reanalysis of saved predictions; it does not introduce new observations or refit or retune the forecasting models. Release: version 1.1.0. Original datasets and their sources are identified in the accompanying documentation.