Backup and disaster recovery
Backup design
The 3-2-1-1-0 pattern and local-first replicas.
How to read status labels: In the pilot (v0.5.0, synthetic data) Simulated in the pilot (no real external system) Designed, not yet built Open decision Not offered / no claim made Planned partner integration
C0221How are backups organized?
By the 3-2-1-1-0 pattern: three copies, two media, one off-site, one immutable or offline, and zero verification errors. Tiers run from a synchronous replica to offline media held in-country.
C0222Is the local database a backup?
The local database is the operational primary and the cloud copy is an encrypted backup replica. This inversion is what allows outage survival.
C0223Can backups be encrypted so the cloud cannot read them?
Yes: bundles are encrypted client-side and signed, so the sync hub stores ciphertext and cannot read them.
C0224Can ransomware delete my backups?
The design gives backup agents write-only credentials, keeps an immutable copy in a separate account and an offline copy with no reachable credentials. Nothing is built yet.
C0225How fast can I recover?
Recovery objectives in the specification are planning targets such as minutes for a single node and hours for a full server from the cloud replica. They are unmeasured and not promised.
C0226What recovery point does a clinic get?
Proposed values only. The specification marks all recovery objectives as recommended and to be validated, and no figure is contractual.
C0227Are restores tested?
One isolated restore drill exists in the pilot, which checks the signature, the digests and the audit chain on synthetic data. Quarterly clean-room drills with a reported pass rate are planned and have not been run.
C0228How is a restore verified?
By signatures, hash chain and ledger anchors, in an isolated environment with malware scanning, then promoted with two-person approval.
C0229Can tampering with backups be detected?
In the pilot, yes in part: the manifest is HMAC-signed and each table has a digest, so an altered backup fails the drill. Anchoring each range on a ledger is designed, not built, and the signing key is demo-grade.
C0230What if the ledger and database disagree after a restore?
Consent and authority follow the ledger; clinical content follows the database with a logged review.
C0231What happens to deleted patients in old backups?
Their keys are destroyed, so immutable ciphertext stays unreadable, and tombstone bundles propagate the erasure.
C0232Where are offline backups kept?
In-country, in clinic or partner custody, with split-custody key escrow. Custody arrangements are not yet defined.
C0233How do backups handle erasure versus immutability?
Crypto-shredding resolves the conflict, subject to legal opinion on whether it counts as erasure.
C0234Does the local backup database sync to another clinic?
Clinic-to-clinic sync is designed where the profile and consent allow, for medium and large tiers.
C0235How is a replica kept consistent?
Gapless sequence numbers, hash chains and replay protection, with retransmission on any gap.
C0236How big is the backup?
Sizing is an open item: structured data is estimated at 5 to 50 MB per patient per year, imaging by modality, all to be measured.
C0237Is there a backup in the pilot?
Yes, in part. The pilot has verifiable backups of the core clinical and identity tables with per-table digests, a signed manifest and an isolated restore drill with dual authorisation to record a promotion. There is no off-site or immutable copy, the signing key is demo-grade, and no disaster-recovery plan has been run.
C0238What about regional disasters?
An in-country disaster recovery site with a 24 hour recovery point target is the design proposal. It is a proposal only.
C0239Is there protection against insider deletion?
Immutable retention, no delete rights for the sync agent and dual-approval purge make single-person deletion impractical by design.