I've felt your pain. Models aren't always trained well enough on edge cases and configs.

Would love to hear how Sup works out for you.