AmShZ's blog

By AmShZ, history, 57 minutes ago, In English

Hello everyone!

After the Repovive Math Contest Grand Opening, we are back with our second live math contest!

This round will be entirely about Combinatorics.

Our goal is still the same: to bring the experience of competitive programming contests to proof-based mathematics. You solve a problem, write your proof, submit it, get a live verdict, and compete on a live leaderboard.

The next contest will be Repovive Math Contest Round 1, held this Tuesday.

The contest is rated for everyone.

What changed after the first contest?

After our first math contest, we selected 534 borderline submissions and graded them manually to create a gold standard. We then ran several experiments to understand how we could improve the judging system.

The accuracies shown below are measured specifically on this curated set of difficult and borderline submissions. The accuracy over all contest submissions is higher.

We also experimented with revised judging instructions and few-shot examples, which significantly improved the performance of several models.

You can view the full study here:

https://x.com/ShayanCJahan/status/2106143879200440800

If you have a background in IOI, IMO, ICPC, or similar competitions and are interested in this kind of work, we would also love to hear from you.

Rules and judging

The rules are the same as in the Grand Opening.

You can find the full rules here:

https://repovive.com/contests/1015/rules

If we make any changes before the contest, they will be announced on that page first.

As before, participants submit written mathematical solutions. Rejected submissions do not add any extra penalty, and each participant may make at most 15 submissions per problem. Your penalty is the time of your last accepted submission, with each minute counting as 1 negative point.

Any use of AI during the contest is strictly prohibited, including for translation. The platform itself can translate problem statements, and solutions may be written in any language.

For this round, the judge will still return a binary verdict: each submission is either Accepted or Rejected. Each solution is evaluated by 5 GPT-6 judge agents.

We do not necessarily consider binary judging the final form of the system. We are actively experimenting with better ways to provide more informative and useful judging in future rounds without revealing information about the solution during the contest.

We learned a lot from the first contest, both from the submissions and from the feedback we received afterward. This round is another step toward making live proof-based math contests work as smoothly and competitively as possible.

We hope you join us this Tuesday!

See you on the leaderboard!

  • Vote: I like it
  • +1
  • Vote: I do not like it