This record provides an EVRPTW benchmark dataset built on the Homberger 1000-customer VRPTW instances (60 problems, six families: R1, R2, C1, C2, RC1, RC2). For each base instance there are two variants: (A) original customer time windows with charging stations appended in standard VRPTW text format, and (B) the same augmented network with distance-based relaxation of customer due dates, limited by the depot closing time. Charging stations are sampled inside the customer coordinate bounds with a minimum separation between stations. A normalized local density score drives heterogeneous attributes: charging rate is set deterministically from that score; price and waiting time are specified via log-normal parameters per station (price_mu / price_sigma, wait_mu / wait_sigma). The package includes augmented .txt files, matching *_stations.csv station tables, params.json (all generation settings and the random seed), ev_specs.json (reference battery capacity Q and consumption rate h for the bundled checks and baselines), README.md, and a code/ folder: Algorithm 1 and Algorithm 2 (regenerate variants A and B), plus scripts for constructive feasibility over several Q values, parallel CON/HC baseline experiments with CSV output, analysis of those results, and an optional sensitivity study of generator parameters. Base instances: homberger_1000_customer_instances/ contains the original, unmodified public Homberger files. All charging-station data, variant B relaxations, and author code are produced by the authors of this dataset.