Author one MAST benchmark item: case stem, expected answer, edge cases, then evaluate any supporting data and pick a target benchmark.