DLM has a lot of flexible features, but it's ultimately restrictive as far as how you can define the content of a fragment: your only choice is content based on the fragment owner's (personal, database-persisted) layout.
We need more options than that. We need DLM fragment content based on pluggable strategies – so the content of a fragment can be the ooutput of a custom algorithm.
This enhancement will allow us to produce fragment content based on things like:
Metadata matching (match user context info to portlet metadata)
Learning analytics data (users like you are interested in the following portlets)
Within the portal JVM, the Java type that encapsulates the nature of a fragment is UserView. We need to convert UserView into an interface-based abstraction (viz. IUserView) and support pluggable concrete implementations.