This seems nice. My use case is more ML driven. Think of it as train test sets but generated by another team/application.