Data registration vs. container

From the definition of the Data Registration, it would seem that the registration itself is the container in which its shapetree-adhering instances are stored. After all, it has a number of metadata predicates (relevant agents, timestamps etc.), but none that points to a container.

Am I right in this assessment, and (if so) why is that the case?

It seems to preclude a division of responsibility, e.g., where I keep my registration metadata on one host, and have it point to my data on another. At first I thought maybe the idea is to have the RS use it for knowing which shapetree a container has to adhere to, but then this would seem to conflict with the idea of "planting trees" from the ShapeTrees spec.

Start by reading the Data Registration definition and the ShapeTrees specification, especially the references to containers, metadata predicates, and “planting trees.” Done means the specification clearly explains whether a registration is itself a container, how data may be hosted separately, and how shape-tree responsibility is assigned.