Write the expected behavior before testing
Choose a limited set of services and document permitted behavior in each state. Distinguish the visible preference, stored representation, transmitted signal and actual service activity. They can disagree even when a button looks correct.
Use fictional information and a clean test state. Record environment, version and any regional or language variations in scope. Do not assume a Google setting controls another provider or that every denied state means no traffic.
Run a sequence from first visit to changed preference
Start without an earlier stored choice and observe the approved initial behavior. Save a preference, navigate and reopen settings. Check that the current display and actual service connection agree on the next load too.
Then change one category while keeping the others. Save and inspect relevant subsequent events. Reopen, make a temporary edit and cancel; cancellation should follow the documented rule rather than accidentally saving unfinished choices.
| Test | Observe | Accepted when |
|---|---|---|
| New visit | Order before first saved choice | Approved initial behavior is respected |
| Next page | Stored value and service signal | Correct state is reused |
| Partial change | Changed and unchanged categories | Only the agreed change takes effect |
| Withdrawn preference | Future events and return visit | New setting governs under the specification |
| Unreadable stored value | Fallback and visible feedback | No silent unsupported permission upgrade |
Exercise delayed loading and storage failure
When services or settings load asynchronously, use a controlled delay test. Check which state a service receives when it starts after a change. A late default response should not overwrite a newly saved choice without support in the model.
Simulate unavailable or unreadable storage in the test environment. Try an old format version if compatibility is required. Confirm both the selected fallback and what the interface claims about the next visit. A withdrawal test does not prove erasure of historical information.
Check the user route and actual service activity
Reach settings using the keyboard and on mobile. A modal dialog needs understandable focus handling on opening, navigation and closing. Assess readable status and working persistence together.
Record initial state, action, observed signal, service activity and next page view. Google Tag Assistant can support checks for relevant Google tags; other connections need their own evidence. Unknown activity remains an open finding with an owner and retest.