All content is held in a shared repository, just like in an online music store, from all different sources and at all different levels of granularity.
The repository maintains a permanent link for all content materials, so you can rely on selected content to be available. It provides versioning, so previous versions of content can be retrieved and changes tracked, and it provides replication to avoid bottleneck situations.
Finally, it provides cataloging data (“metadata”) for the content, both static (title, keyword, subject, taxonomy, …) and dynamic (usage, context, accesses, difficulty, …) – data that will feed into the recommender.
Will this actually work? We know it does, since it does in LON-CAPA, where over the years over 450,000 pieces of content have been contributed to a common repository. This content is going to form the base of LON-CAPA.