The indexer machinery should be able to use embedded versions of assets, preferably from a tiered set of sources: Live/Database, Filesystem, Embedded.

This would ensure there's some data in all cases, even if it's horrifically old.