Follow durable progress
Operational actions submit durable jobs. Record the returned operation reference and inspect the progress/history screen. Closing a browser tab does not own the job's lifetime, and a queued request is not a completed host change.
Review the actual final state and service evidence. Failed, blocked, compensated and uncertain outcomes need different recovery decisions. A capability-unavailable response should remain unavailable; it is not a reason to fabricate a resource or repeatedly resubmit.
Use the correct role
Admin has installation-level tools subject to current grants, MFA and freshness. Customer sessions see only their account's resources. An administrator's View session is audited, expiring and read-only, with a visible banner and exit action; it does not reveal the customer's password or allow ordinary writes.
If a sensitive action requires identity verification, complete the supported password/MFA flow, review the current request and submit explicitly again. Do not assume a pending form was automatically retried.
Check effects after interruption
Preserve the operation ID, time and sanitised correlation reference. Inspect current resource and service state before deciding whether an existing operation can resume. Use the matching release's recovery workflow rather than clearing journals or starting another potentially duplicate remote action.