According to your reply the information on 'site' will be available. Therefore will be possible to know whether a subject belongs to one of the site already available in the training phase. Moreover it will be possible to know whether two subjects belong to the same unseen site or they come from different sites.

A further question concerned with the release of test data. The dataset of BradleyHospital/BrownUniversity will be released as part of the test set. For this portion of data is known in advance the distribution of the category labels. Since you mentioned that a further dataset from an unseen site will be released as test set, it would be important to know whether the distribution of category labels will be published as well.