Hello everyone!
Have you ever wished that math and combinatorics problems had the same live contest experience as competitive programming?
We have, so we are building exactly that on Repovive!
After beta testing our new Math section, we are ready to host the Repovive Math Contest Grand Opening.
You can already try our beta math problems here: https://repovive.com/problems/math
The theme of our first contest will be Combinatorics. The round will contain 5 combinatorics problems to be solved in 2 hours and 30 minutes. The problem set is intended to be suitable for a wide range of levels, so everyone is welcome to participate.
The contest is rated for everyone.
Since this is the first large live test of this format, we would especially love to have you participate, try the judging system, and share your feedback afterward.
- Start Time: Tuesday, 29 September 2026, 14:35 (UTC)
- Duration: $$$2$$$ hours, $$$30$$$ minutes
- Problems: $$$5$$$ combinatorics problems
- Points: $$$500,\ 1000,\ 1500,\ 2000,\ 2500$$$
- Registration: https://repovive.com/contests/math-round-1
Prizes
We have an $800 prize pool for the grand opening!
- 1st place: $250
- 2nd place: $150
- 3rd place: $100
There are also additional prizes:
- First Accept: $25 for the first accepted solution on each of the 5 problems
- Random Top 100: 4 participants from the top 100 will receive $25 each
- Random Participant: 3 participants from the contest will receive $25 each
Rules
- This is a mathematical problem solving contest. No programming is required.
- You submit a written solution for each problem.
- Your penalty is the time of your last accepted submission. Each minute counts as 1 negative point.
- Wrong submissions do not add any extra penalty.
- You may make at most 15 submissions per problem.
- Your goal is to get as many problems accepted as possible within the contest.
- Any use of AI is strictly prohibited, including using AI for translation.
- The platform can translate problem statements into different languages for you.
- You may write your solution in any language. The judging system supports solutions written in different languages.
How judging works
Instead of a traditional code judge, your written mathematical solutions will be evaluated by 5 GPT 6 judges.
You can write your proof and reasoning directly in text. The judges check whether your submitted solution correctly solves the problem.
Judging is completely binary. There is no partial score. Each submission is either Accepted or Rejected.
LLM based judging for mathematical reasoning has also been studied academically. For some background on the strengths and limitations of this approach, you can read the paper From Calculation to Adjudication: Examining LLM Judges on Mathematical Reasoning Tasks.
This contest is also an experiment for us. We want to see how well a live competitive format for mathematical problem solving can work, both technically and competitively.
We would really appreciate your participation and feedback.
See you on the leaderboard!
UPD1: Congratulations to all prize winners!
$250 first place: yusa kimura (fact493)
$150 second place: Kosuke Kaneshita (potato167)
$100 third place: 1434
$25 First Accept prizes
One prize for each problem:
- Problem A: Костя Тыщук
- Problem B: hos-lyric (hos.lyric)
- Problem C: Kosuke Kaneshita (potato167)
- Problem D: yusa kimura (fact493)
- Problem E: Kosuke Kaneshita (potato167)
$25 Random Top 100 prizes
Four winners:
$25 Random Participant prizes
Three winners:
That's $800 in total prizes. Random winners were drawn from all contestants who submitted during the round.
Congratulations to the winners, and thanks to everyone who took part in our first Math contest!
Winners, keep an eye on your email inbox for instructions on receiving your prize.
Do you plan to add lean support?
It would be interesting if, rather than judging on freeform proof, you first provide freeform, you receive lean translation and confirm/clarify intent, then that is judged.
Yes! Lean support is currently under implementation and should be available in the future.
Will written submissions be on paper or typed in latex?
Typed.
You can write your solution in any language and any format you prefer, including plain text or LaTeX.
Hello, can this contest time be changed to a weekend? Preferably after the starter round on Sunday or another Sunday.
This contest has already been scheduled, so we probably won’t change its time.
For the upcoming contests, though, we’ll definitely try to schedule them on weekends whenever possible.
what do you think about changing the grading to IMO/Putnam style (either 0-7 or 0-10 scale) instead of natural language feedback? I feel like natural language is giving too much information back to the users.
We may change both the binary grading and the grading scale in the future. This is our first math contest, so the rules and judging system may change quite a lot over time based on feedback.
Also, the feedback currently shown in the beta is more detailed than what participants will receive during the actual contest, so it should not reveal as much information.
The main reason we currently prefer binary grading is fairness. With something like an IMO-style 0–7 scale, we would need a much stricter grading protocol, and there could be variance between graders that makes the results less fair.
If we become confident that our judging system can grade accurately and consistently on such a scale, I think it could be a good option.
Thanks for the suggestion!
Isn't the feedback on rejected submissions a pretty significant bug? I just tested it and a WA submission gave me a specific mathematical hint identifying the flaw in my proof. Since wrong submissions have no extra penalty and there are 15 attempts, this effectively lets contestants use the judge as an interactive proof debugger: submit → get a hint → fix → resubmit
This seems quite different from normal contest judging and could potentially make some problems much easier
Take a spanning tree (T) of (G). It has (n-1) edges, so there are
[ m-(n-1)=m-n+1 ]
edges outside (T) Choose any subset of the edges outside (T) For each chosen edge (e=(u,v)), add the unique path from (u) to (v) in (T) This gives a subgraph where every vertex has even degree: the added edge and the path together form a cycle Thus, the set of edges with even degree is completely determined by which edges outside (T) we choose Once those are fixed, the tree edges are forced: starting from the leaves of (T) their inclusion is uniquely determined by the requirement that every vertex has even degree So every subset of the (m-n+1) non-tree edges gives exactly one valid (S) and every valid (S) is obtained this way Therefore the number of such subsets is
[ \boxed{2^{m-n+1}} ]
Take a spanning tree (T) of (G). It has (n-1) edges, so there are
[ m-(n-1)=m-n+1 ] edges outside (T) Now choose any subset of the edges outside (T). The edges of (T) are then uniquely determined by the requirement that every vertex has even degree: start from the leaves of (T), and for each leaf, its edge to its parent is forced by the parity of its other incident edges. Remove the leaf and continue Thus every subset of the (m-n+1) non-tree edges gives exactly one valid (S), and every valid (S) is obtained this way. Therefore the number of such subsets is
[ \boxed{2^{m-n+1}} ]
With 15 attempts and no WA penalty one could deliberately submit progressively incomplete proofs and use the judge to locate the missing pieces.
You are right to point this out.
What you are seeing here is the training mode. During live contests, the judge will not provide these hints on rejected submissions, so it cannot be used as an interactive proof debugger in this way.
We will also make this distinction clearer on the platform.
Interesting, with AI help now we can have math contest auto judge.
Is it allowed to write program by hand to investigate some property of the problem? Or only pen and paper?
No. As in standard math contests, you are not allowed to use programs or other computational tools to investigate the problem. You should solve it using only your own mathematical reasoning.
Of course, this is not something we can always detect, so it is partly an honor-system rule between us and the participants. However, if we find clear evidence that someone used such tools during the contest, we may disqualify that user.
Also, in your proof, you cannot justify a claim by saying that you checked it with code. Every step of the proof should be mathematically justified and follow logically from the previous steps.
I couldn't find this restriction anywhere on the official rules page. Could you please explicitly add it there?
Since Repovive is also a competitive programming platform, I don't think it would be obvious to participants that writing code to explore properties of a problem is prohibited. It would be helpful to have this restriction clearly stated in the official rules.
Sure. Thanks for noticing.
I confirmed that I can get AC using Japanese in the beta version. This isn't against the rules, right?
No, that's ok.
Wait, y'all actually cooked
Is it allowed to google an unfamiliar definition and nothing else? I might not know the meaning exactly because I don't use that definition in life at all but I can know it in other words. Or I would like to try solving the problem after learning the definition. One more question: can it be added to type formulas in Typst format (https://typstpad.com example)? They are really convenient.
PRO tip: if you cant get accepted, try translating into other languages (dont do it in actual contest tho)
Just for fun, try getting the shortest possible accepted solution to P1 and P3.
bipartite checkerboardTrivial checkerboardThe goals are
9 characters
5 characters
Please announce the random winner of Repovive Premier Round 7 — $300 in Prizes, Repovive Premier Round 8 — $300 Prize Pool and First Math contest.
Because there is a very long queue in your prize announcement, I think.
Thank you
Am I correct in understanding that a separate rating, something like a “math rating” rather than the CP rating, will change?
Yes, math rating and rank is completely separate than CP.
If a proof gets rejected in the live round, is it just a no, or does it cost something on the scoreboard the way a WA does on Codeforces? The beta hints being off is already clear. I didn't see the penalty part.
There are no penalties on wrong submissions
Overall, I am satisfied with the contest and appreciate the interesting experience. However, I have some thoughts:
Felt the same... overall great contest enjoyed it but getting AC even when your ideas are correct was difficult.
Can we have the leaderboard?
This contest is an awesome idea! I enjoy it a lot.
Would there be official editorial be published (or hint at least)? I didn't find anything on the page.
(upd. nvm, I found it at the problem section instead of contest section)
This is going in the history books. I can't wait for the next one!
The leaderboard for y'all (you can also see problems from the dashboard and the solutions from the leaderboard by tapping the scores of the user you wanna see solution of without registering): https://repovive.com/contests/math-round-1/ranking
This is very very cool! I ran into a case where the AI was stupid though (and suspect that it might be common?):
For B on the opening contest contest I submitted the following:
Every box can remain except for the center one when n is odd. Suppose that box i is not the center box and WLOG suppose that it is to the left of center. We can make it the remaining box by applying the following operations: for all j from 1 to i — 1, merge j into j + 1. Then for all j from n to i + 1, merge j into j — 1.
Conversely, suppose it is the center box, and that it remains till the end. WLOG, at some point it must be the leftmost box with more boxes remaining to its right. Since the total money is conserved, and no money could’ve moved from the original left half to the right half, box i must have >n/2 money inside, and thus will get removed in its next operation.
This got rejected for the following incorrect reason:
The proposed sequence of merges is not always legal, since it sometimes transfers tokens from a box with more tokens into one with fewer; the converse also does not justify its claim about the center box’s token count
And got accepted after I added the following:
This sequence of operations is legal, because it always moves money from a box with more money into a box with less money, as specified in the problem statement…
I didn’t do the live contest so I’m not sure how the feedback works, but I think that it’s possible that quite a few submissions got rejected not because of errors/lack of detail but just cus the judge hallucinated?