I feel like CodeChef is too generous when it comes to rating increase during provisional rating calculation. While it may be intended to place a higher performing candidate into their correct division as soon as possible, it also ignores the similar performing candidate which has his/her rating assigned (no longer provisional) getting a meagre rating increase.
Looking at the standings it was clear getting a rank of ~10 in division 2 gives one +137 rating increase with people going easily above 2100+ whereas the same guy at some rating like 2030 goes to 2050.
So my point is → If someone wants to increase his/her rating its far better for him/her to make new accounts to gain rating than sticking to their single account. This is strictly against moral competitive programming grounds. I feel like the provisional rating system needs a rework. Please CodeChef look into this.
For experienced users, the new system rewards consistency. You might consistently be at a particular level, but do terribly in one contest, and the system won’t punish you with a huge -ve delta for it. It will be treated as more of an outlier, and it’ll give you more chances, so to speak. Similarly, just a single great contest won’t spike your ratings by a lot. If you consistently do better than what the system expects you to, then it’ll give you higher +ve deltas.
Just see the difference between the two rating graph tabs to get an idea of the difference in the systems. The old system is very chaotic, while the new system rewards you for consistently improving over time.
If the 2030 participant gives a couple more good contests, the rating will jump up faster. You can read more about the model, here. We do not plan to change this.
While I do agree with the consistency remark and the need for consistent contest performances according to one’s level, the point that I made still stands.
The point is → The rating increase after rating has been assigned works as you mentioned (about the consistency part) and even I think it doesn’t need a rework. The rating increase before rating has been assigned (provisional phase) is the one that needs rework. You need to cap the amount of delta increase people are getting in the first 5 contests.
Maybe introducing something like: Once you get to Div1 you no longer get huge delta increase can be done. Because right now anyone can get to 2100+ if they make a new account and perform well in the first 5 contests. Doing so after rating has been assigned is a little tedious.
We agree that the initial phase is not ideal. We have iterated on this multiple times, and this is one of the better models we have after running experiments which balances multiple opposing factors. We will probably get back to this at some point again, but don’t plan to do so right now.