Aggregate supporting data for Dominance of the eligibility rule over model choice in warning of clinically significant hypoglycaemia

Observed aggregate numerical results supporting the manuscript Dominance of the eligibility rule over model choice in warning of clinically significant hypoglycaemia. The archive contains 39 CSV files with 12,641 aggregate rows, a data dictionary and dataset metadata. It covers baseline comparisons, eligibility cutoffs, calibration, warning mechanisms, forecast diagnostics, five training seeds, calibration-rank composition and frozen external-cohort evaluations.Only author-derived global, cohort-level, training-seed, histogram, threshold-sweep and bootstrap summaries are included. Third-party CGM records, participant identifiers and predictions, temporal traces, event vectors, individual demographic values and fitted model weights are not redistributed. Primary-data access remains governed by the original providers.The registered source-event regret training objective and multiseed horizon-averaged primary bootstrap were not executed. Source-bootstrap summaries are descriptive. The completed participant-bootstrap contrasts are single-seed and horizon-specific. External evaluation supports the composition rather than an external replication of the eligibility reversal; HUPA-UCM is a descriptive stress test.This work was supported by the National Science and Technology Major Project of the Ministry of Science and Technology of China (No. 2024ZD0531200), followed by the National Natural Science Foundation of China (Nos. 82302842 and 82560451). Code: https://github.com/gxCaesar/cgm-hypoglycaemia-warning. Author-derived aggregate results are licensed under CC BY 4.0; this does not relicense the underlying third-party data.