Review changes before moving on
Compare the implementation to the plan and check nearby failure cases.
By the end, you can inspect a change, try expected and edge-case paths, and report what remains.
Step 1: Ask what counts as success
An is an observable result you agreed to: the Save button fits at 390px; typed text survives a rejected save; A cannot read B’s notes. A confident summary does not establish these outcomes.
Step 2: Use both the preview and change list
The preview shows behavior. The diff is the list of code changes. A text edit can look right while an unrelated query changes. Removing a page filter is a reason to investigate; it is not automatic proof of a leak because database policies may still enforce access.
Step 3: Request evidence for each check
Ask the AI which check it ran, the observed result, and what it could not verify. Require a phone-width interaction, keyboard access, invalid-input recovery, and ordinary-account data tests where relevant. Do not accept line count or a passing build as proof of these behaviors.
Step 4: Keep corrections small
Try the main flow and nearby failures using fictional data. Preserve useful work and reject unexplained scope growth. what passed, failed, and remains out of scope before asking for a focused correction.
Compare the diff and preview against these checks: button fits at 390px; rejected save preserves input; keyboard submission and written errors work; another account cannot read private . For each, report the action and observed evidence. Explain unrelated edits and state what you could not verify. Illustrative before / after (behavior, not ready-to-run code): - Copy edit plus unexplained removal of own-user query filter + Copy edit retained; query change reviewed; ordinary-user access verified
A screenshot proves a visible state; it does not prove every interaction, saved result, or permission rule.
Sources and further reading
The AI says a mobile save fix works. Your requirements were: the button fits at 390px and entered text survives a rejected save. What evidence should you ?
Choose one answer.