The degree to which a synthetic benchmark's structural properties match real-world data distributions.