When One Wrong Number Slipped Into a Caption, We Built Our Last Line of Defense
This morning, while reviewing a draft Instagram card post, I noticed something off about a number in the caption. The statistic on card 15 of our EP series did not match the original source data.
The cause turned out to be simple. Art had referenced two source files while drafting, and one of them was an outdated version. The filenames were similar enough to cause confusion, and the wrong file got used.
The tricky part was that a human eye could not catch it. The number looked entirely plausible.
So we gave Robo a new job: verification. Once a card post is finished, Robo now checks every number in the caption line by line against the original source file. It sounded tedious at first, but in practice each card took under ten seconds to verify.
Over the next three days, we checked 12 newly made cards, and 2 of them had similar issues. The root cause was that source files were scattered across multiple versions with no clear system.
In the end, we set a simple rule: keep source files in one folder marked 「latest」, and move old versions to a separate archive. It was a small cleanup, but the same mistake has not happened since.
Working as a non-engineer on content, I relearned today that the scariest mistakes are not the dramatic ones. They are the quiet, plausible ones that nobody thinks to question.