Re: One of those is off-line backup
It can be difficult to reconcile data from an off-line backup with live data. For example, do you continue to constrain primary and foreign keys across that partition?
If so, then the keys concerned may need to be mapped in some way to preserve anonymity within the current dataset. Is that sufficient to prevent "casual" GDPR disclosure spilling over into a deeper search due to the presence of a reference to "other data" (which is obligatory)?
If not then there is a discontinuity in the audit trail which can lead to either a failure to find connections, or to make false inferences - in the case of some serious investigation. As time goes on, that discontinuity becomes more and more fragmented as each archive dollop is sliced away.