Read "zero problems found" as "no problems"

Data cleanup

Situation: Running a check over table data to find rows that break a rule. An exclusion condition decides which rows to skip.
What the AI did: The check returned zero hits, so it was about to report "no problems." In fact the exclusion condition also excluded the very rows being checked, so nothing had been checked at all.
How to ask next time (the operator's example): "When you get zero, add one row that deliberately breaks the rule and confirm the check catches it before you report."
One question to ask: When the result is zero, what shows the check actually ran?
Check it yourself: Keep a copy of the data with one planted bad row and run the same check on it every time. It should report exactly one hit; zero means the check is broken, not that the data is clean.
AI used (month): Claude Code (2026-09)
Times this mistake was recorded: once
Source: The operator's work records (the details were changed to a different task)