Although the evaluation set remains private to avoid contamination, we detail our methodology below. The Benchmark is already used by some frontier models.