I think all of the top 10 participants have an equal chance of winning this competition. It's even possible that others who are not currently in the top 10 could end up in the top spots on the Private Leaderboard. Most of the competitors in the top 10 are aware of which consumer_device entries are included in the Public Leaderboard and which ones are in the Private Leaderboard. Only 3 of them appear in the Public Leaderboard, while the majority are in the Private one. So, if you tune your model based on the Public Leaderboard, there’s a high risk of overfitting.