Hi trev, nice work on the competition so far! Thanks for the questions.
I won't go into how exactly the data is divided (just to keep things fair). I'll just say that we mixed a couple strategies to make sure it was representative of how the algorithm would be used in the future. That is, the new data coming in will never look exactly like the data you already have.
The visible public leaderboard is not the final score (so that people can't just overfit what they see there). It is calculated on a subset of the test data. I think the explainer for this is buried in the rules--we'll make sure to call that out more prominently.
