What is an approval workflow? Separate decisions from execution results
Explain requests, approval states and subsequent execution, checking timeouts, rejection and duplicate notices.
Understand the concept
Approval workflows collect authorized decisions and continue accordingly. Notifications and reads are not approval, and approved actions can still fail. Define request versions, approvers and states, preserve decisions and verify execution separately. Models must not guess or grant human authorization.
Decision and execution states need different evidence
Power Automate creates approvals, waits for responses and branches to actions. Distinguish waiting, rejection, approval and execution failure. Decisions need a specific request version and rules for later edits. Buttons or model judgments do not replace records. Define multi-person counts and order rather than assuming one response completes all conditions.
Use simulated requests to test rejection and timeout
Use public simulated requests with approved and rejected responses and check bounded test actions. Timeout must remain pending or terminate by rule, never imply consent. Duplicate notices should reference one request. No approval, contact or business operation occurred here. Inspect target state so a stored decision does not hide failed execution.
Version people and policy changes
Replay after role or field changes and verify authority sources. Preserve decisions and execution IDs and check prior effects before retries. Include follow-up and exceptions. AI may organize material without expanding authorization. Link request content, decision evidence and actual outcome and clearly retain undecided cases.
Checked sources
These sources support the concepts and methods described here. Examples and test scope are labeled separately; check the version of the tools you use.