QI-Bench-100 v0.1: A benchmark of 1,000 labelled global quantum-network topologies for evaluating learned surrogates

QI-Bench-100 is a benchmark of 1,000 labelled global quantum-network topologies built over 100 real cities, with heterogeneous link types (fibre, satellite, ground station). Each topology is labelled by an exact Werner-fidelity-optimal analytical oracle for four physical objectives: fidelity, entanglement rate, latency and coverage. This artefact accompanies the manuscript "QI-Bench-100: Benchmarking Graph and Tabular Surrogates for Global Quantum-Internet Topology Evaluation". It contains the dataset (topologies_v2.npz, 1,000 topologies), the full source code (dataset generation, exact analytical oracle, channel models, the HeteroGNN-Heavy-Mean architecture and the training pipeline), the reported per-seed results, and the trained checkpoints for every reported seed. The split used throughout the manuscript is generated with seed 42 and the three reported model runs use seeds 0, 1 and 2, both fixed in the code so results reproduce exactly as published. Data are released under CC BY 4.0; code, results and checkpoints under the MIT licence.