Withdrawing approval should prevent eligible future execution, but it cannot erase an external effect that already committed. Track the decision and execution lifecycle separately so staff receive an accurate result.
Withdrawal changes trusted decision state
Approval withdrawal revokes an earlier decision about a specific proposal. The workflow needs to establish whether execution has not started, is running, completed, or has an unknown outcome before choosing the recovery path.
For a proposed design, bind approval to proposal revision and record withdrawal through an authenticated authorised actor. The executor checks the current decision at a defined checkpoint before committing the effect. A stale notification saying “approved” must not override trusted current state.
That checkpoint alone cannot cancel a remote action already in flight. Use the destination’s supported cancellation or reconciliation mechanism, and define any separate compensation process. Avoid promising that a withdrawal means the original action never happened.
Test a fictional customer-update withdrawal
A manager approves changing a customer’s service address, then discovers a mismatch and withdraws the decision. If the job is still queued, it should become ineligible for execution. The staff view should show that the update was stopped before any write.
Now move the withdrawal after the external service accepted the update but before its response arrived. The system should mark the outcome uncertain and reconcile. Creating a new reverse update without checking could overwrite a later legitimate staff change.
| Withdrawal timing | Required behaviour |
|---|---|
| Before execution claim | Approved work cannot start |
| After claim, before final checkpoint | Executor observes current decision |
| During remote write | Supported cancellation or reconciliation |
| After confirmed completion | Separate authorised correction if appropriate |
| Late stale approval event | Cannot revive withdrawn permission |
Use a simulated timeline and controlled service responses. Inspect the authoritative customer record and operation history, not only the approval screen.
Make the boundary clear to staff
A useful response distinguishes “withdrawn before execution,” “withdrawal requested while running,” and “action already completed.” Staff need to know what remains possible and who owns recovery. A single cancelled badge can hide materially different outcomes.
Withdrawal adds coordination and race handling. Keep the policy understandable, and choose a freshness check appropriate to the actual external contract. For irreversible actions, the review should explain the point after which withdrawal cannot guarantee prevention.
This workflow is proposed. The voice integration case study describes related backend work without claiming this approval mechanism. Pair it with pending-action cancellation.
Approval withdrawal FAQ
Does withdrawal always undo the action?
No. It revokes permission for eligible future execution; committed effects need separate handling.
Can the interface alone enforce withdrawal?
No. The execution boundary must resolve current trusted decision state.
What should a race test prove?
The final decision and actual effect remain consistent or explicitly uncertain, with no unsupported claim of cancellation.