Backup Cadence for the Sovereign Stack: Restore-First Thinking
Backups are not copies; backups are promises. The sovereign stack’s promise is that nothing is lost — the vault, the graph, the posts, the customer data — and the only way to keep that promise is to test the restore, not just take the copy. Restore-first thinking inverts the classic order: design the restore first, then make the backup serve it. A backup that cannot be restored is a storage bill, not a promise.
The three-legged stool
Backup discipline rests on three legs: frequency (how often the state is captured — the cadence must match how much loss is acceptable), integrity (the backup must be verified — checksummed, tested, restorable), and location (the backup must survive the failure it is protecting against — off-box, off-site, in the vault’s own redundancy). Docker Swarm persistence (S7.9) documented how the stack survives restarts; the backup cadence extends that to survive disasters.
Restore-first, in practice
The restore-first drill is a scheduled operation, not an emergency: quarterly, take a backup, spin up a throwaway environment, and restore into it — the database, the media library, the configuration. The drill measures restore time, finds missing pieces, and updates the runbook (OP-1). The drill is the health check for the backup itself (OP-8): a backup that has never been restored is an unverified claim. The cost governance layer (OP-4) allocates budget for the drill, because the drill is insurance, and insurance is a line item.
Backups as a product promise
For customers of the sovereign stack, backup discipline is a selling point: the migration service (EX-8) and the managed services (EX-12) promise continuity, and continuity is only credible if the restore is tested. The audit layer (S7.15) can verify the drill records the way it verifies agent actions. Restore-first thinking turns backup from a chore into a capability — and a capability is something the stack can sell.
Grounded in the OP operations series, the S7.9 swarm-persistence article, the OP-1 runbook article, the OP-4 cost article, and the EX managed-service examples. Ninth article in the Round D operations track.
