Supporting data for AI-driven cislunar GNSS channel modelling from lunar observations

Author-derived data supporting the manuscript AI-driven cislunar GNSS channel modelling from lunar observations. The dataset contains one-minute LuGRE link-state features, training/validation/test partitions, fitted residual-learning models, model outputs, evaluation metrics and numerical source data for the figures and tables. Supporting analyses cover persistent and link-state-dependent learned residuals, grouped model-input attribution, antenna-reference sensitivity, temperature sensitivity, leave-one-operation-out evaluation and early-operation calibration. The accompanying documentation records data provenance, variables, units, original public-data sources and verification results. Core model prediction/refit, temperature-refit and chronological-calibration results have been reproduced; the release does not claim a fully verified one-command reconstruction of all figures from raw observations. Original third-party observation archives, orbit/attitude products, SPICE kernels, antenna workbooks and paper PDFs are not bundled. The archive includes processed reference grids and digitized curves with source attribution and source-specific rights notices. CC BY 4.0 applies to the authors' original contributions, not as a relicensing of third-party material. Companion code: https://github.com/SJTU-GNC/LuGRE-npj-wireless-technology.