I would appreciate an official clarification from the organizers. I have seen similar situations before, and they can easily lead to misunderstandings about model performance. While waiting for an answer, I might take a short break and play Veck IO before returning to the competition. Clear documentation always makes these projects much easier to follow.