Data and code for: Accurate graph neural network surrogates can mislead green infrastructure placement

Companion data and code for a manuscript on simulator verification of graph neural network (GNN) surrogates for green infrastructure placement on two urban drainage networks in Suwon, South Korea. Includes design-of-experiments allocations and node flood labels, node-level siting envelopes, surrogate-optimization and EPA-SWMM verification outputs, greedy frontiers, sensitivity and rainfall results, trained models, and analysis scripts. Node coordinates were removed. The raw municipal sewer GIS, network edge tables, and SWMM input models are not included because of the data provider's redistribution terms and infrastructure sensitivity, so the deposit supports inspection and reanalysis of reported results but not a rerun of the simulations or surrogate training. See README.md.