A nice step since new participant's ratings constantly drop.

The idea is great, atcoder has been successful in reducing multiple accounts. Great initiative.

Seems like a great update! Will the rating field in the CF API display the visible rating or the backend one used to calculate changes? I’m curious about whether this will break rating change prediction tools until their creators update them to account for this change.

Additionally, will the visible rating or the true rating be used to evaluate eligibility for Div 2/3/4 rounds? For example, suppose a new user jumps to 1700 true rating (thus, 800 display rating) after one round: can they participate officially in a Div. 3 round afterwards?

If new participants are discouraged in the starting due to lower rating points, then if they get negative delta in their first few contests then their rating will be <500, which will be much more discouraging.

I suggest initial rating to 500, in practice as well as during calculations. Instead lets balance rating inflation/deflation, so that that for the first few contests rating inflation/deflation is low, and then it gradually hits a stable value with more contests. I hope you get the general idea.

На аткодере рейтинг идет с нуля. Кажется что это еще лучше решает три обозначенные проблемы

While reading the rules I thought I am reading some codeforces A problem statement and I need to answer the final rating of a newly registered account.

What about the old users who had participated in less than six contests but at least one contest participation? Do their rating change?

You can see my rating graph, it is a clear example of harsh rating losses for newcomers. It was a palindrome at the beginning!

Really happy to see this, you're right, I was also tempted to make another account, but I didn't lose hope and today I am 1500+ :D

This should have happened a long time ago!

Codeforces should look to Atcoder for inspiration for its rating system.

Connect the codeforces account with respective country's UID (Unique Identity) Number. Like there is AADHAR NUMBER in India.This might reduce number of fake accounts.

First of all, thanks for this step !!

I have a question in mind . What is the approximately maximum value of d1+500 + d2+ 350 + ...+ d6+ 50. If this value is always less than 1400 then there is no problem .

But , Suppose this value exceeds 1400 in the 4th round ( d1+d2+...d4) . and suppose the participant's rating is 1500 . Then still he will be treated as 1400 rated and will cause serious rating instability.

What about graph of rating?

After these changes are done.

Apparently, a similar idea is used on AtCoder, but I don't know the details.

It would make sense to research prior state of the art before creating one's own implementation

Will that new person participate at rated contests based on his/her displayed rating or based on the real rating?

If the former happens, one can get a really high rating from participating at Div3 and Div2, which is not fair.

Suggestion: is it also may be possible to make the first few contests(say 3) unrated for the new users. This will help the new users to actually get accustomed to the style of contests here and will decrease the number of fake accounts

like div1 + div2 rounds will we see div2+div3 or div3+div4 in the future because it looks like in future div4 contests frequency is going to increase?

Can anyone briefly explain what Mike has written? I'm too stupid to understand the technical intricacies involved

The upcoming Div.2 round is the last chance for us, new reds to create as many alts as we can in order to take part in the following Div.1 contests with ease

There will be one absurd thing about this though, if you don't solve any problem in your first contest, your display rating will still increase by around 350. I believe that would look absurd.

I believe it would make more sense to keep the display rating locked at 0 until a user has solved at least one problem in an official round (and then add all of the pending "promotions" in rating at once or maybe delay the promotions).

I think 6 contests to get to your true rating sounds fair, and more is definitely unnecessary. AtCoder's system is a bit unfortunate in that the rating seems to climb really slowly, which means mine still hasn't converged after 13 contests (I'm 100 points lower than my contest performance would indicate). It's mildly annoying and I'd prefer to avoid that.

Just for fun, last week I made a comparison of rating vs. percentile across different competition platforms. Notice how AtCoder's median is 165 (yes, mid-grey) which doesn't look reasonable to me.

(X-axis is exponential, the data might not be 100% accurate, CodeChef includes all users and not only active).

But I suggest the beginning imaginary rating be 1500 instead of 1400. So as you said new accounts will contribute less to rating. The existing specialists and experts will find more difficult to gain rating. So someone who got 1700 before the start of this rule got is much easier than someone will now.

Will this be also applied for those who already registered but didn't participate in more than six contests? Or will it be applicable for only those who will create an account after the implementation of this?

Rating is just a number. The main raiting — is your knowledge. (but you still should not look at my rating history)

How it's done on AtCoder:

$$$rating_{shown} = rating_{true} - f(competitions)$$$, where $$$f(x)$$$ is a magic function that starts at $$$1200$$$ and tends to $$$0$$$ as $$$x$$$ tends to infinty.

What you did is equivalent, but with $$$f(1) = 900$$$, $$$f(2) = 550$$$ ... $$$f(7) = 0$$$.

@Mike, I liked your idea of calculating the rating. One thing that I would like to share with you guys — Like if a guy has a rating of 1599 and participates in Div 3, goes to 1700+ , guy with 1600 can also do the same(not comparing any one). Can this be taken into picture "IF YOU CROSS A DIV — CAN IT BE LIKE Min(DIV X + 50, new Rating). Ps — Just a THOUGHT. (No personal issues).

We may see marked increase in participation in the upcoming Div2 round as people who are confident enough might not want to wait for 5-6 round to get to their deserved rating. Also better start at 1500 than 1400 if you feel you stand somewhere in between.

Great initiative!

This may be quite unrelated to this post, but I think this adds another reason to suggest this now.

I think we need an up-to-date version of this 9-year old FAQ. A big portion of this post contains old information which doesn't correspond to how the system works nowadays. It was already too much out of date when I first joined CF, and now it has even more complicated features like Div. 3 & 4 and this modified rating system. Even though I really like this new rating system in general, this may make newcomers a lot more confused about their rating changes if they can't find this information easily.

Maybe it would make more sense (in that the system might be cleaner) to rethink the entire rating system (that includes round divisions etc) rather than to add features to the existing system.

It is really good idea.It would definitely stop multiple accounts formation and also would be nice to boost up confidence of new participant

If i have less than 1600 rating, but more than 1600 true rating (1400 + $$$d_1$$$ + $$$d_2$$$ + ...) . Is div3 rated for me?

So basically now you can get to master just by hitting top scores on 3 div3s in a row. Or maybe div4s (why on earth is that even a thing when it was what div3 was meant to be). The whole system just keeps getting more and more broken lol.

If someones actual rating becomes more than 1600. Just give him his actual rating instead of waiting for 6 contests. The current system will just increase inflation with 2200 and 2300 users/alts participating in div2s. Becoming GMs using div2s only.

Very good step.

Assigning contests based on displayed rating? This is opening up a whole new can of worms here. If it ain't broke, don't fix it.

Грамотный подход

It sounds perfectly correct . This will definitely demotivate to create new accounts as they create when their ratings drops so much. Best Luck to participants for the next contest rounds. Wish all high rating!

Yeah but can you wait for like 3 more div 2 contests ? Gotta farm those filthy rating and get yellow before deflation hits.

Oh,no new accounts become Master in just 3 rounds then.

I don't know if it's a good thing,in my opinion it's more likely to be,but time will tell.

Hope high rating for new comers

Apparently, a similar idea is used on AtCoder, but I don't know the details.

That's a great idea to give the new accounts more confidence. For me, it's really terrible if you work really hard but your rating just drops.

I like the way it works in AtCoder, there my rating went down the first time after 7 contests. Assuming it will work here similar it is a good change.

Still the problem that top 10 of Div3 contests make a huge jump in rating, but that is another story.

And I very much apreciate that new accounts become less atractive.

Мне кажется странным и больше вносящим непонимание, чем реально помогающим. В плане в Atcoder я не решил ни 1ой задачи но все равно получил +37. Но за что? То есть вместо того, чтобы дать более-менее реальную оценку сразу есть какие-то непонятные алгоритмы (это вот сейчас вы объяснили, а пользователи, которые не в теме будут только недоумевать). Более прозрачная и простая система гораздо лучше в плане фидбэка, чем вот такие ухищрения. Можно изменить начальную точку и считать не от 1400, а от 1200 например и это кажется более разумным. Потому что рейтинг перестает быть trustable, а становится системой манипуляций, которому ты не доверяешь как пользователь, а это с моей точки зрения более важно, чем позитивный фидбэк в плане отрицательного роста. Можно сделать чтобы ты получил рейтинг нужно иметь 3 участия как калибровка и только тогда показывать рейтинг и это кажется идеей более хорошей.

Hey..i am a school student and i just got introduced to codeforces. i took the above Div 3 challenge yesterday and managed to solve 2 problems, post that my laptop got shut unfortunately. However, both the submitted questions were accepted in the first test as well as post the open hack and system test session. But i haven't got any updates regarding my performance or ratings. I am new to this. Please tell me what am i supposed to do?

People with new account are becoming master and candidate master. That does not seems to be fair imo. This could promote people to make new account.

Where can I get the comprehensive information about how the ratings are calculated in CodeForces by the way?

MikeMirzayanov I think there are few bugs in the new rating system as many people who gave their first contest yesterday have become candidate master with just 2-3 problems solved.

I have observed that problems in CF Round #644 (Div 3) were quite easy. Except the last, all problems had 2000+ successful submissions during the contest. Due to this, the rating of participant in 1400-1600 exploded. Adding more to this, new accounts became masters just by solving 2 problems last night. LOLzz.

This might be due to new rating system. However, this completely destroys the motive of the new rating mechanism, that was to reduce creation of new accounts.

I propose two points after seeing what happened:

1. Withdraw the new rating system.

2. Make prev. round unrated for participants who were in 1400-1600 before the contest. As the problems were indeed suitable for Div 4 contests only.

Comment if you agree/disagree with my proposal.

Maybe it's a good choice, but we may need time to adapt to it

bruh wtf im mad bro sir i lost my exopert im sad bro

MikeMirzayanov I think there are still a few problems look at Droping_points rating. He/sh's solved all problems but the rating was not enough added.

I have given seven contests till date and in each one my rating has only fallen, even though ive solved at least two problems in each contest(apart from one). Its quite disencouraging tbh. Still waiting for even a slight increase in my rating.

I'm a bit confused with the rules apparently. Will the ratings for other users decrease if starting a new calculation from zero?

I think with the fast growing number of users, even 1400 will still make the rating inflated at a super high rate. As for most newbies, I think a starting number of 1000 or less is more appropriate(It doesn't affect the rating showed anyway)

Guys I was curious about one thing, but I don't want to spam Codeforces with a new blog so asking here... I am seeing some high-rated guys from my country being online (almost) continuously. They aren't submitting anything most of the time, not even in practice(except in a few days). So what are they doing? Is this some kind of browser connection bug? I have no interest in their life, just want to satisfy my curiosity

Just remove div3 and div4.

I wish Codeforces's own formula.

I like Codeforce the most. So I disagree with following someone else's formula.

There is no need to change the formula.

What is the point of this update? I mean, as far as I remember, CF uses some kind of Elo rating system (one of Elo, Glicko, Glicko-2, etc.), where the starting rating is just an additive constant which does not affect any computations (just because 1500 rating looks prettier than 0). So, in fact, this update adds 100 additional points to all old profiles, and therefore breaks whole rating system, because these numbers mean nothing now (literally nothing, +100 points for reds is not the same as +100 points for blues). And moreover, it does not even solve the problem of one-contest profiles, just moves it 100 points lower.

Btw, AtCoder uses its own crazy rating system, which is definitely not from Elo family, so everything above does not apply to it.

Besides the implementation, I agree that current rating system is not the best for CF purposes and maybe should be changed, but it requires a lot more research.

When I started comp. programming I suffered from the same issue. I feared my rating would drop and didn't participate in contest, rather solved problems later. Although, I now understand that rating is just a depiction of your skills and their was no point of doing what I did, I really appreciate this step and it will be really in making sure that the ultra-beginners don't get demotivated in the starting itself.

Just do what we have in chess. No rating for first 10 contests, and after the 10th contest you get your rating = average perfomance in 10 contests.

Before that, when N < 10 contests have been played, show approximate rating = average perfomance in N contests in the profile.

Perfomance is "if you had this rating, it wouldn't change after the contest".

Guys, we need to update the СF predictor, because it will show the wrong rating.

Where have the comments gone?

_BeginnerCoder replied to your comment

Don't think it is a good idea!!!

Separation of the real rating and the displayed rating does not seem right to me You could just drop the initial rating to 1000 or 0 or whatever would solve the problem

The current system is much better.

For me when i started codeforces, I had almost no idea about cp and knowing just some basics of C I was in the same position of rapid rating loss, but I think those initial rating drops made me work harder and improve myself, had it been the new rating system I might not have given my best because I HAD NOTHING TO LOSE, reason for that may be: even if I perform bad my rating will increase initially.

I don't mean disregard the new method by any means :D

Советую тогда на каждом аккаунте сделать такой первоначальный пересчёт рейтинга. Чтобы у всех было красивое растущее выравнивание, а не только у новичков. Кому-то правда придётся урезать рейтинг, но я думаю, вам, как организации, это простительно, причём он урежется лишь на первые 6 контестов. Мне кажется, так будет лучше.

Can anyone tell me why we new users are not assigned 0 rating? So if he solve a single question he will get motivated.

Apparently "/ratings?order=BY_RATING_CHANGE_DESC" isn't interesting anymore.

Great initiative though.

[deleted manually]

why there is no strict action against plagiarism.

Good idea.

For (not that applicable) comparison, AOE2:DE uses several ranked ladders such that if you have good rating in one, it's easy to get good rating in another in a few games, and it requires you to play 10 games before you're visibly ranked — before that, only you can see your hypothetical rating.

FYI, I finally made a decent write-up of my contest rating algorithm (at least the first 8 pages of it): https://github.com/EbTech/EloR/blob/master/paper/EloR.pdf (2021 update: https://arxiv.org/abs/2101.00400)

Here's how everyone on Codeforces ranks according to it, including the performance and rating change of their most recent participation. Max ratings are in parentheses: https://raw.githubusercontent.com/EbTech/EloR/master/data/codeforces/CFratings.txt

The displayed rating is $$$1500 - 2(\sigma - 100)$$$, where $$$\sigma$$$ starts at $$$350$$$ and approaches $$$100$$$ as members become more experienced. So, the displaying rating is initially $$$1000$$$. That's a "penalty" of $$$500$$$, which shrinks to about $$$207.9$$$, $$$120.3$$$, $$$76.6$$$, $$$50.9$$$, $$$34.6$$$, $$$23.8$$$, $$$16.5$$$, $$$11.5$$$, $$$8.1$$$, and $$$5.7$$$ after competing ten times.

My system is a rigorous generalization of Elo/Glicko to the case of large numbers of simultaneous participants. It reduces inflation (but doesn't eliminate it), reduces the influence of new members, limits the size of rating jumps for experienced contestants, and produces (I believe) a more reasonable distribution of ratings (see the table here for a comparison). The math does all this naturally, without special hacks.

As a bonus, I recently optimized the implementation so that the entire history of Codeforces can be simulated on my laptop in 25 minutes. I haven't run many experiments or detailed comparisons, but I'd gladly assist anyone who wants to try!