I've been reading the latest post by Mike about a new feature of non-rated registration, and upon reaching the following sentence a horrible realisation struck me down.
During the round, you will not be able to see whether another participant is registered as rated or non-rated.
This essentially makes predicting the rating delta during the contest an impossible task. Even now the predictions are not accurate, but this change, I imagine, will render them utterly useless. The way it looks right now is that Carrot will no longer be useful.
If you're not yet familiar with Carrot, I recommend you try this wonderful extension out while it still works.
Edit: Although, now that I think of it, Carrot's function of evaluating the rating your performance corresponds to will still be quite handy.
Auto comment: topic has been updated by smalyarovsky (previous revision, new revision, compare).
Well they'll probably follow through with it, and a good amount of people have been asking for it, but I think it's a sad decision to add. If I bother with round I want it to be against highest percentage actually trying, I want to be able to be able to guess how good my performance actually is in contest, and I don't want unrated people clogging up queue.
As far as I know it was already broken a little after the fake rating system (where new users start at 0 but are secretly 1400 in the system and gain "free" rating for the first few contests) was added. At the time Mike said he would provide trueRating api which would allow a program to get the true rating of a user but apparently it was never added and therefore carrot just assumes displayed rating is true rating and usually underestimates rating changes.
I don't think unrated registration will be used particularly often so it shouldn't change predictions much, people (including me) like to see the stupid number go up. Even then, as long as it's not used like significantly at some rating and never at another rating, it should balance out the errors it incurs.