Hey,
It depnends on scenario, but there is many use cases that hack data
change from somebody with admin privileges could be disaster.
That is the place where data history could come with help. Some basic
solution would be trigger which writes previous version of record
to some other table. Trigger however can be disabled or removed (crazy
solution would be to provide pernament
triggers and tables which can only be pernamently inserted).
Then we have also possibility to modify tablespace directly on disk.
But Postgres has ability to not override records when two concurrent
transaction modify data to provide MVCC.
So what about pernamently not vacuumable tables. Adding some xid log
tables with hash of record on hash on previous hash.
I think that would be serious additional advantage for best open source
relational databes.
Best regards,
Marek Mosiewicz