How dependant is the second round of the competition on the first one

I was wondering how strongly the first and second round of these competitions are coupled? I am assuming that you have to exactly discuss the model + techniques that gave you the best score on the LB, right?

It could be important to clarify this further, since there are often trade-offs between model applicability/interpretability/… and the performance you can gain from it (e.g. taking averages of submissions, or ensembling different models together is very deleterious for the interpretability and applicability of your model)