Currently, we rely on Python's pickling format with several unoptimized dictionaries. We should migrate this metadata and probably have an index over their title ID and UID (if determined necessary).