Public Evidence Release: Robust Quantile, Safe Polytope, and Certified Learning Augmentation for FNM-RSMA

Overview This dataset provides the public simulation evidence, numerical verification records, and figure-reproduction resources for a study of certified robust quality-of-service (QoS) optimization in downlink Rate-Splitting Multiple Access (RSMA) systems over Fluctuating Nakagami-m (FNM) fading channels. The underlying approach combines robust outage quantile construction, a certified feasible power-allocation region, and learning-augmented numerical optimization with verification and fallback mechanisms. Contents The release includes frozen simulation results across five random seeds, three FNM/SIC parameter regimes, and four user configurations (2, 3, 4, and 6 users), including in-distribution and out-of-distribution evaluation contexts. The archive contains aggregated results, confidence intervals, public baseline configurations, numerical equivalence-audit records, figure source data, SVG/PNG visualizations, and Python scripts for figure regeneration. Principal Findings The expanded simulation reports a mean online solver speedup of 1.1516× (95% confidence interval: 1.1178–1.1853) and a mean end-to-end pipeline speedup of 1.0212× (95% confidence interval: 1.0174–1.0251). The confidence interval for the paired absolute pipeline-time difference includes zero; therefore, universal end-to-end runtime superiority is not established. The recorded out-of-distribution fallback rate is 16.7%. Reproducibility and Scope The released files support independent inspection of the archived numerical results and regeneration of the published figures. The private production optimizer and experiment-orchestration implementation are not included; consequently, this archive does not provide a complete end-to-end rerun of the original experiments. The reported evaluation is simulation-based. No physical-layer hardware deployment or experimental validation on a real communication system is represented. Licensing Research data, figures, and documentation are distributed under Creative Commons Attribution 4.0 International (CC BY 4.0). The included public figure-regeneration scripts are distributed under the MIT License.