Accurate prediction of materials phase diagrams from first principles remains a central challenge in computational materials science. Machine-learning interatomic potentials can provide near-DFT ...