Intermittent connectivity

Plan for connection loss as part of the Customer-managed Azure design. An entitled application can use a local Edge Gateway for ordinary governance calls and keep its central VeriProof endpoint configured for validated handoffs.

Identify which connection can fail: the application’s connection to VeriProof, or a VeriProof service’s connection to an external dependency. The affected workflows differ.

Questions to resolve

• Which work can continue without a new policy decision?
• Which actions must fail closed?
• Which telemetry can the SDK retry without duplicating a committed side effect?
• How long can evidence remain pending?
• Which portal and review workflows require live service access?
• How will operators detect backlog growth and failed delivery?
• Which proof-provider operations require outbound connectivity?
• What recovery evidence proves that queued work was processed once?

Supported SDK behavior

Set VERIPROOF_EDGE_ENDPOINT to the local gateway address. The gateway uses signed policies and an authenticated governance profile. It does not silently reroute a failed protected action to another destination.

The gateway journals accepted evidence and stop acknowledgements before reconciliation. Keep its journal volume across restarts, and confirm backlog delivery after central connectivity returns.

Ordinary governance can continue only while the signed policy lease and governance profile remain valid. The gateway rejects new work when required state expires or protected storage is unavailable.

Never treat a network error as permission to continue a protected action. Follow the selected SDK’s policy and failure contract.

Qualification plan

Test at least these conditions in the target Azure architecture:

1. endpoint unavailable before a governed action
2. connection loss during a request
3. telemetry retry after a successful server commit
4. application restart with pending work
5. proof submission backlog and recovery
6. human-review pause and callback recovery
7. monitoring, alerting, and operator response

For each test, record what was accepted, what remains pending, and whether the business action ran. A recovered connection alone does not confirm recovery of the work.

Use Edge Gateway setup, SDK troubleshooting, and the Customer-managed Azure guide as the starting points.