Code and derived data for explainable road selection at 1:50,000 and 1:100,000 in the Netherlands

This deposit contains the analysis code and derived data for a road-selection experiment in eleven areas of the Netherlands. The processed road graphs include segment attributes, network features and retention labels derived from the September 2026 Kadaster TOP10NL, TOP50NL and TOP100NL data. The archive contains fitted GraphSAGE weights for five initializations per configuration, study-area splits, segment-level predictions, input attributions and validation-based knowledge selections. The included scripts can reconstruct the selection metrics, connectivity checks, paired attribution-feedback comparison, and decision for a 24 m street in Friesland. Source geographical data: CC-BY Kadaster 2026 (CC BY 4.0). The study-area layers, labels and network features are derived products.