GatekeeperQA：無障礙檢查結果消失不代表已確認修正

A finding disappears from the next scan. Can your report call it fixed?

Not yet. The element could have moved, the page could be incomplete, or the rule could now require manual review.

We build GatekeeperQA. While auditing its retest workflow after community feedback, we found a mismatch: individual-finding retests correctly reported uncertainty when a later scan had an unresolved check for the same rule, but the bulk comparison could count that finding as “no longer detected.” We corrected the bulk result and added regression coverage.

One example, three different conclusions

Consider an earlier label violation on a form field. These are illustrative scenarios, not customer results:

Later scan	Safe conclusion	Next step
The same finding is present	Still detected	Inspect the fix and retest
The rule has an unresolved manual check	Uncertain / not verified	Review the affected control manually
No matching finding, with comparable completed coverage	No longer detected	Verify that the same control remains and works

An unreviewed check must stay unreviewed. A missing page result or execution error cannot establish a passing comparison. “Fixed” needs evidence appropriate to the issue, not just an empty list.

Match the finding, not its row number

Rows can move between reports. One approach is an exact-location identifier built from the page URL, rule and selector structure. It stays consistent when rows are reordered.

That is not a permanent DOM identity: a changed selector, different URL or replaced element still needs investigation. Preserve the original report references instead of assuming two similar-looking rows describe the same control.

Give the next reviewer enough context

For a useful handoff, record:

The earlier and later report IDs and scan dates.
The tested page and finding location.
The automated comparison result, separate from human review.
What a person tested, which input method they used, and what remains untested.

Store structured report references and dates with attached retest evidence, and carry them into exports. Do not infer dates or version information that older text-only notes never recorded.

Regression tests should cover row reordering, unresolved checks and incomplete page coverage. Passing those tests is software regression evidence, not proof of accessibility conformance. Keyboard, screen-reader and other relevant manual checks remain necessary.

A regression case worth keeping

Given an earlier label violation and a later unresolved label check, assert that the bulk comparison reports uncertainty instead of counting the finding as no longer detected. Test the individual result and the bulk summary against the same fixture so they cannot quietly disagree.

Disclosure: this article was generated with AI and checked against the project implementation and test evidence. The example scenarios are illustrative, not firsthand user-testing results.