Trust Center
Architecture: pilot versus design
The pilot runs in the cloud with an on-site hub for outages. The designed system adds sealed on-site servers, a ledger of hashes and hardware hardening. The diagram marks which is which.
Solid elements run in the pilot. Dashed elements are designed and not built.
The pilot Pilot
- Electron client and a tablet web build, over HTTPS and secure WebSockets.
- A Cloudflare Worker with one Durable Object per facility for live fan-out and timers.
- One database with a clinic identifier on every tenant table (test-enforced).
- Additive-only migrations (28 so far).
- An on-site hub on Node 22 and SQLite, built from the same worker code, with signed and encrypted replication to the cloud.
- A server circuit that live-tests every module after each deploy.
The design Designed
- A sealed, cabinet-sized on-site server per clinic as the operational primary (the on-site hub software exists; the sealed hardware and primary-copy role do not).
- A federated ledger per clinic or network holding only hashes, consent and audit events.
- Three backup copies on two media with one offsite and one offline (the pilot has verifiable backups and an isolated restore drill only).
- Server-to-server transfer in three streams with hybrid post-quantum key exchange.
Safe degradation ladder
From normal operation down to full local-only. Downtime mode follows this ladder (L0 to L5) in the pilot. Some thresholds are open decisions.
Partly met today
The specification requires local operation without the network. The on-site hub and downtime mode were built to meet it and are tested on synthetic data, but the pilot’s main hub is in the cloud, a replication conflict has no resolution screen, and no site drill has been run.