In the problem statement of DIVLAND in APRIL15, it says
Your solution will be tested against only on 20% of the test files during the contest and will be rejudged against 100% after the end of competition.
From my understanding this means a submission will potentially receive changed verdict due to the new tests (e.g. from AC to WA). However, I originally thought (from my previous experience in long contests) that a submission will be run against all test cases during the contest, but only showing a partial scores. Therefore I made a clarification request to the problem author of DIVLAND and received a reply saying that
It’s running against all test cases, but revealing score only to 20%.
However I found one of my APRIL15 AC submissions received TLE after rejudging. This is probably due to my program trying to run up to the TL with slight tolerance :). This indicates that all submissions were indeed re-run during rejudging, making solutions running up to TL more risky.
So I was kind of confused by how the challenge problem is judged and rejudged. I tried but didn’t find a clear explanation on the site regarding this. I believe the better way is to run against all tests (otherwise you could get WA after rejudging :|). But I think re-running solutions is not necessary for rejudging purpose, which could undesirably change AC to TLE. We can preserve the full scores without showing them during the contest, and just update the scores after the contest.
