Wouldn't it be practical for translator and project manager alike to be able to distinguish between those 100%-matches that were made from scratch and those that were post-edited from NMT?
It would help to recognize the potentially more direct translations…