Hi, Codeforces!

Remember this post?

If not, let me remind you. For the last 3–4 days, I have been building a browser extension for Codeforces that shows your pure rank without cheaters.

How does it work?

The idea is simple:

1. The extension gets the Codeforces contest standings.
2. It goes through each handle.
3. It makes a prediction using an ML model based on the user’s performance in previous contests.

Some of the features include: skipped contests, unusually fast rating deltas (+100, +200 after previously gaining only +10 or +20), rating behavior after the first 5 contests when the rating becomes more stable, and other signals. You can check the full list of features in cheating_dataset.csv, which I used to train the classifier.

Interestingly, this extension may also be useful for finding suspicious users after a contest, since the model seems to be fairly accurate.

Examples

Now let me show some examples:

Let’s take a user with a high score. The higher the score, the more confident the model is. For example: studyyyzz.

Let’s check this user:

Maybe the model only catches obvious cases? Let’s take a user with a lower score. For example: HubRis504.

As you can see, it works quite well.

FAQ

• How can I install this extension?
  To install the extension, check this post.

• Is the model always correct?
  No. This is an ML model, and ML models cannot be 100% accurate by nature. Please do not accuse people just because the model predicted that they might be suspicious.

• Can I improve it?
  Yes, of course. It is open-source. You can check the implementation and improve it here: click