[jerven]

Not sure KDE went back to a relational model. The main change I feel was going from a single central database (Virtuoso) which had a copy of everything to keeping the data where it is as much as possible and only store copies where needed for performance.

Virtuoso being an impressive DB is not the most stable or resource use friendly datastore for desktop use.

This decentralised data storage actually makes a lot of sense. And I hope to work on something similar for life science data except that the unifying API will be SPARQL instead of a C++ API. (Not to say that a single C++ API does make a lot of sense for the KDE project, where it does not for life science data.)