Problems span several mathematical subjects and difficulty levels, with answers represented as free-form mathematical text. Automated checking requires normalization or symbolic handling, and some cases benefit from human adjudication. Results should distinguish answer correctness from the quality of generated reasoning.