Hello,
The use-case would be WS projects where the pre-translation is done against usual translation memories, AND translation memories that contain raw machine translations. The idea is that translators can see the translations coming from the "machine translation memory" marked as AT and not as full matches.
Thanks,
Daniel