Uptime and offline operation
What happens when the internet drops
Honest current behavior versus the designed local-first behavior.
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
C0198What happens in the pilot when the internet goes down?
On the cloud hub alone, a network outage stops alerts and messaging; the app shows a red offline banner with a reconnect countdown and disables sending, and missed alerts and messages are delivered on reconnect and flagged late. Where an on-site hub is installed, it keeps alerts and messaging running locally and replicates to the cloud when the link returns. That has been tested on synthetic data only, and a replication conflict has no resolution screen yet.
C0199Will the production design work without internet?
In part, and the pilot already does some of it. The on-site hub stays operational for alerts and messaging when the WAN is down, and downtime mode adds banners, sealed downtime reports and back-entry. The sealed on-site server as the operational primary is the remaining design. No uptime figure is promised.
C0200Can I open charts while offline in the pilot?
Only read-only and only what was already open. The browser keeps an offline read cache in memory, which is read-only and is cleared when the screen locks. Downtime mode adds sealed downtime reports and back-entry of paper records afterwards. This site claims no more for full chart use on an on-site hub than what has been tested: alerts and messaging.
C0201Is working offline safer from hackers?
The design states that being offline improves security against remote attack, which is why a limiter trip pauses only sync and leaves local care running.
C0202What is the safe-mode ladder?
Six levels from normal operation through WAN down, ledger down, partial server, server down and site lost. At each level care continues locally as far as possible, and cross-site features are reduced first.
C0203What is a downtime emergency data set?
In the pilot it is smaller than the design: a read-only cache held in browser memory and cleared on lock, plus sealed downtime reports. The design adds current census, allergies, medications, code status, emergency contacts and last vitals refreshed to two encrypted local appliances. Code status is derived from order names because no structured field exists, and the appliances are not built.
C0204How long do cached credentials work offline?
The design example is at most 72 hours since last sync, logged locally and reconciled later. Example value, open for decision.
C0205Are paper downtime forms part of the plan?
Partly built. The pilot generates sealed downtime reports locally, re-verifies their SHA-256 on every read, and supports back-entry of paper records with the original and entry times. Locked, tamper-sealed, shredded-on-refresh paper and per-shift refresh are design.
C0206How often are downtime drills run?
The design calls for downtime drills with clinical staff twice a year per unit and quarterly restore drills. These are plans, not performed drills.
C0207What uptime percentage do you guarantee?
None. Availability figures in the specification are planning targets to be validated by measurement, and no service-level agreement is offered. The availability promise per tier is an open decision.
C0208Is there an uptime SLA for the small tier?
No. The small-tier support model is undecided; best-effort business-hours support is the working default and not a commitment.
C0209What is the weakest structural area?
Availability under attack. Systems coupled for security are also levers for denial of service, so the design uses quorums, local autonomy and a non-cryptographic emergency mode.
C0210What if the single cabinet fails?
At the small profile the cabinet, its power and its uplink are single points of failure. The mitigation is an uninterruptible power supply, downtime mode with sealed reports and the cloud copy, with no claim of high availability. The sealed cabinet and power equipment are not built.
C0211Does a power cut stop care?
The reference profile includes an online UPS sized for at least an hour at small-clinic load and a generator interface where available. The clinic supplies its own electricity.
C0212What is the dead-man policy?
If the server loses contact with its key authority for a configured number of days it refuses to unseal after restart, with an emergency unseal by quorum so care is never endangered. Interval is an open value.
C0213What happens to queued messages after an outage?
Queues are persistent and drain in order on reconnect, with backpressure and an alert instead of silent drops.
C0214Are late alerts labelled?
Yes in the pilot: alerts delivered after reconnect carry a late flag so staff know they are not live.
C0215Can the local network work if the server is fine but the internet is down?
Yes: alerts, messaging and audit are designed to operate entirely on the local network with no cloud push in the alert path.
C0216How is a site rebuilt after total loss?
The design rebuilds from the cloud replica or an immutable vault into a clean-room environment, verified by signatures and chain, then promoted with dual authorization. The pilot has an isolated restore drill with signature and chain checks, and promotion records the dual-authorised decision only; it does not swap the live database. A yearly rebuild drill is planned, not performed.
C0217How does the system behave during a cloud outage?
With an on-site hub the clinic is meant to carry on: alerts and messaging keep running on it and replicate to the cloud afterwards. That is tested on synthetic data only. A conflict, such as a duplicate record created on both sides, pauses sync until a person resolves it, and the resolution screen is not built.
C0218What happens when one server node fails?
Medium and large profiles use replicas with automatic failover. Small clinics run one cabinet, with a second downtime appliance recommended.
C0219How is the health of the system monitored?
Metrics, PHI-free logs, queue depth, backup age, certificate expiry and device health are designed, not built. The server circuit live-tests every module after each deploy, which is a release gate and not continuous monitoring.
C0220Who is on call for the small tier?
Undecided. Severity-based support is planned: 24x7 for medium and large, business hours for small unless a fee is agreed.