Data and code for: Universal machine-learning interatomic potentials reduce magnetostructural energetics to elasticity: a benchmark on antiferromagnetic CrN and MnO

Reference data (with per-row provenance), input structures, raw model outputs, derived statistics, relaxed geometries, analysis and scoring scripts, pinned environments and figure generators for a zero-shot benchmark of seven spin-blind universal machine-learning interatomic potentials (MACE-MP-0, CHGNet, SevenNet-0, M3GNet-MatPES, MatterSim, Orb-v3, GRACE-2L-OMAT) on the magnetic energetics of antiferromagnetic CrN and MnO. No new DFT calculations were performed; all reference values are taken from published work.