Rolling labelled data updation

What might be the hurdle(s) in publishing the actual labels for the rounds that are already resolved? This will add a new dynamic to the approach.

The hurdle is they don’t want you to have that information and will add a dynamic they don’t want to occur, namely overfitting to the most recent data.

But that will give the users flexibility to choose the training data (old or new), which will result in variability of submissions, and it ultimately benefits numerai.