Suppose a data scientist purchased a license to use a particular data set. Then they must go through a publisher to get the actual data set. It might be stored on Azure. The publisher would hold the Azure access keys, and would do the checks necessary to ensure that the data scientist has access before letting them get the data set.
Publishers should run Brizo and a Keeper node.
Initially, there won’t be many Ocean marketplaces or publishers, so most marketplaces will also act as publishers, and most publishers will also act as marketplaces. Therefore, for now, see the docs for marketplaces.