The test data for this problem has been reworked. New tricky tests have been added. At the same time the majority of the old tests have been removed.

From now on, the total number of tests in this problem is limited to 20 tests. It means if new good tests are discovered, they couldn’t be simply added but should replace some of the old ones.

If you’re stuck, don’t spend time on figuring out the test data and hyperoptimizing, because the test could be replaced with a similar one any time. Instead, detect that your heuristics fail and launch another one in this case. There is no universal test in the test data that makes all the heuristics fail at the same time.

All solutions have been rejudged. Only a few authors still have AC after the rejudge. Because of test removal some of the previously rejected solutions got accepted.

Edited by author 03.05.2026 22:50

Who created the new tests? Is there a publication?

As far as I understand, the old ones were created as part of this work: https://is.ifmo.ru/disser/buzdalov-dissertation.pdf

Edited by author 04.05.2026 18:18

Edited by moderator 14.05.2026 04:00

Some of them are my tests. There is no publication. I didn’t have access to other people’s solutions, so the tests are fairly general. LLM wasn’t used when creating the tests.

Edited by moderator 14.05.2026 04:00

Are there any new, previously unknown theoretical foundations behind the new tests (and your new quick solution), or is it simply a rethinking based on new experience or a fresh perspective?

Edited by moderator 14.05.2026 04:01

To be honest, it’s hard to say; the idea for these tests came to me a few months ago when I decided to improve my old solution, so overall, it’s a fresh perspective. But they do have a theoretical basis, more so than not.

Edited by moderator 14.05.2026 04:01

There are huge gaps in my search area—the solution contains a very large number of magic constants, which, however, were optimizations for the old tests; without them, it still passed in terms of time. I expected that if new tests were selected, a TLE would almost certainly occur. But since the tests were generated without access to the solutions, it’s clear why there’s no TLE with the new tests.

Edited by author 07.05.2026 19:16

Edited by moderator 14.05.2026 04:01