A Connected App Is Not a Verified Workflow
In this fictional example, an AI-assisted workflow copies office supply requests into a shared order queue. It stopped when its connection expired. After the team reconnected the app, the workflow said it was ready. That does not tell the team which queue will receive new requests or which requests from the outage already made it there.
Keep automated processing paused until the owner approves the checks and any replay. Use fictional data in an approved test destination. If no safe test is available, record that the workflow was not verified.
1. Confirm the Account and Destination
Check the trigger source, connected account, destination, and allowed actions against the approved setup. Two workspaces may show queues with the same name. Confirm the workspace and queue, not just the label. Do not broaden access to get past an error.
Try it: Check an AI automation for permission scope.
2. Test One Request and Check Both Sides
Before the run, write down what should happen: test request TEST-04
should produce exactly one row in the test queue and nothing in the live
queue. Compare the run log with the actual destination. Confirm that the
expected row arrived and check the live destination through an approved
read-only view or ask its owner to confirm. An AI summary of the run is
not a substitute for either check.
Try it: Check an AI incident timeline against an event log.
3. Reconcile Missed Work Before Replaying It
Use a stable request ID to compare the backlog with the destination. Mark each item as already written, missing, or uncertain. Do not replay items already written. Review missing items with the owner before replaying them, and hold uncertain items until there is evidence of what happened.
Try it: Check an automation before retrying.
Example: The Run Failed, but the Request Arrived
The log shows a timeout for REQ-018, but the intended queue already
contains that request. Do not replay it. REQ-019 does not appear in
the queue, but the team cannot yet verify whether a pending run will write
it. Hold that item. Ask the workflow owner to check pending runs
before processing it again.
Leave a Note for the Next Reconnection
Record the tested destination, evidence, backlog decision, and who will check anything unresolved. Decide separately whether scheduled processing can resume and whether each backlog item can be replayed. Use your team’s approved storage for the record, and leave out secrets and private request data.
Download the reconnection checklist (.txt)
The practice packs use separate fictional datasets. Start with one check or return to the practice packs to choose another.