At certain sizes it will be difficult to read every relationship back out of spicedb through the API. If you plan to do manual meddling in the database, maybe a resync-solution based on change data capture would fit the bill? For any kind of wholesale reconciliation loop I would be very worried about race conditions between the process of reading all of the data, computing the reconciliation actions, and applying them, running concurrently with the interactive changes from the user.