I cannot imagine the labor costs involved in flagging rejections and then fielding questions from Merchers asking for further information about every little thing that gets rejected.
The new uploader already has a filter that detects the words "shirt," etc., so why not incorporate a similar mechanism to flag anything that is going to trigger a manual review? I've recently had a couple of rejections for words that are not now and never have been in violation of policies but the robot thought otherwise. Why not just let me know in advance that the system will identify the contraction who're as being the same as whore? A system that does that could save us all a LOT of time and money.