Why is the final list result different from the historical record?
Like all DrivenData competitions, unless otherwise stated there is a “private” portion of the test set and a “public” portion of the test set. The final scores are based on the private portion from which you do not get feedback during the competition in order to avoid overfiting. The results of this competition will be finalized once all of the submissions complete their execution and we review the submissions of the winners.
First, if there is a default rule, it is recommended to explain it in the game interface, which is unfair to newcomers.
Second, the number of submissions does not exceed 10 times, and it is hard to say that it is overfitting for such a large amount of data competition.
Third, all our submissions have results on the AB test set. According to the rules you said, the final result is not the last submission, but the best history on B. According to the results, it is more like overfitting. B test set.