Remove Virtual Contestants from Contest Leaderboards

Revision en1, by Geothermal, 2020-05-28 21:06:51

As most of you are likely aware, cheating in virtual contests (typically by submitting prewritten or copied solutions in order to appear at the top of the leaderboard) is very common. Most recently, wannabecandidatemaster submitted solutions to all problems in this round in order to take first place in the standings, and bemandrei competed virtually in this round after competing in the round officially, and ended up in second place on the leaderboard.

In response to the rise of VC cheating (as well as some other issues with the VC system I'll describe below), I'd like to make a simple proposal: Remove all virtual participants from the contest leaderboards. The one exception is that you should see your own name on leaderboards for contests you VCed (that is, virtual contestants should be visible only to themselves). Similarly, virtual participants should not be counted in the solve statistics during future contestants' VCs.

Cheating on VCs is harmful for two reasons. First, it makes it impossible to review the scoreboard as it appeared at the end of the round; in particular, this makes it difficult to discern accurate rankings for out of contest participants. Though this may seem like a relatively minor issue, Div 2, 3, and 4 rounds regularly attract hundreds or even thousands of competitors too high rated to compete officially, so it has an exceptionally broad scope and affects a large number of users.

Second, and more importantly, VC cheaters damage the VC experience for future competitors by making the solve statistics inaccurate. Many competitors use the dashboard's view of how many people have solved each problem in order to make strategic decisions during contests. In the round when I first reached GM, I achieved a high rank by noticing a few early solves on D, correctly discerning that it was unusually easy, and solving it before B or C in order to achieve a better penalty time. This kind of strategy is very common, but the presence of VC cheaters makes it much less practical because in pretty much every contest, when there are a few early solves on the hardest problems, they're not because those problems are easy--they're the result of VC cheaters. The result is that VC cheaters make the VC experience less realistic, so removing them from the solve statistics and leaderboards is critical to achieving the intended goals of the VC feature.

One might claim that the solution is simply to identify cheaters and remove them from the standings after the fact. However, this is a worse approach than removing VCs from the leaderboards altogether for three reasons. First, this approach requires continual effort on the part of the site administrators to remove VC cheaters, whereas my proposed solution needs only a one-time update. Second, it is impossible to accurately detect cheaters, especially because many forms of cheating are less obvious--for example, VCers can look up the test data using a second browser in order to gain an advantage when debugging, and there is no way to accurately distinguish between this and simply guessing the error quickly.

Third and finally, VCs are fundamentally different from participating in a contest live, so even in the absence of cheating, they should not appear on the leaderboards. There are a number of reasons for this, but I see two of them as most important. First, VC solutions are judged on all tests, rather than just the pre-tests, so it is impossible for them to FST: if their solution fails a test not originally included in the pre-tests, they get a chance to fix their error and resubmit, giving them a huge advantage over in-contest competitors. Second, frequently, Codeforces users release blog posts discussing techniques that came up in recent rounds; competitors who read an article inspired by a certain round and then VCed that round would have an advantage over anyone who did that round live.

As a result, I claim that the ideal way to deal with VCs is to remove them from the leaderboards entirely. Please feel free to leave thoughts below; I'm happy to discuss this further with anyone interested or to answer questions about my proposal.


  Rev. Lang. By When Δ Comment
en1 English Geothermal 2020-05-28 21:06:51 4379 Initial revision (published)