An approval can remain valid while the target integration is offline, but execution still needs current prerequisites and accurate outcome handling. Test the decision lifecycle and the external effect separately under controlled outages.
Outage tests need several failure windows
Approval-outage testing checks what happens when a dependency fails before execution, during a request, after a possible commit, or while status is being reconciled. A generic unavailable exception covers only part of that risk.
For a proposed fixture, retain proposal revision, reviewer decision, relevant resource version, operation identity, and fake integration state. Control when the outage begins and ends. Use a simulated clock for expiry and retry deadlines.
Do not infer that a timeout cancelled the approved action. The remote system may have committed it. Likewise, restoring connectivity should not automatically revive an expired or withdrawn proposal.
Rehearse a fictional approved address update
A supervisor approves the current address correction. The integration rejects the request before execution because it is offline. The operation can remain pending under the approved retry policy, while staff see that the record is not yet changed.
In another fixture, the remote write succeeds but its response is lost. Recovery must reconcile the existing operation before repeating it. During the wait, the caller corrects the address again, so stale-intent protection also applies.
| Outage timing | Required invariant |
|---|---|
| Before approved write begins | No false completion |
| After approval expires | No new execution under stale permission |
| After possible remote commit | Reconcile before retry |
| During withdrawal request | Current decision state preserved |
| After service restoration | Only eligible current work resumes |
Inspect the fake target records, decision history, and staff status. A workflow eventually completing is insufficient if it briefly applied the wrong address or duplicated a customer notification.
Include recovery ownership
When automatic attempts end, create an actionable review item with known effect state and safe evidence references. Staff should know whether to clarify, repair credentials, reconcile, or await an integration owner.
Outage testing increases fixture complexity, but it directly exercises the boundary where approval systems often overpromise. Keep a small critical suite for these windows and expand it when an incident reveals a new mechanism.
This test plan is proposed. The voice integration case study describes relevant workflow and backend work. Pair it with tool outage testing.
Approval outage FAQ
Does approval mean the action will definitely complete?
No. Integration availability and current execution checks still matter.
Should an outage extend approval automatically?
Only under an explicit authorised policy that preserves the decision’s meaning and freshness requirements.
What proves recovery is correct?
The final authorised state is verified, stale work remains blocked, duplicate effects are absent, and unresolved cases have owners.