I have built backup into my applications. I backup to flat tables from the database. If there is a recovery, the source data is available to rebuild the database. I have the database schema defined in the program for recovery of tables.

The recovery should be a controlled process. I would not want the users to arbitrarily select a table from a backup (directory copy) and overright a bad table -- there may be missing parent or child records (the parent table might be corrupt but the child table is not).