This competition is a 3 track one, and the overview recommends that we use placeholder to represent both the unavailable IDs and tracks we are not participating in.
However, the leaderboard sort by the mean for the 3 tracks, leading to two issues.
1) it will be difficult to know the actual state of progress per track.
2) competitors could just be duplicating predictions for one track into others they are not participating in.
Can the host look into this? As the point of the leaderboard is to see how competitive our performance is compared to others.
Evaluating this is going to be a mess as competitors just paste scores of the biggest model everywhere.
That was unfortunately unavoidable 😂