SymptomThe routine runs to completion and reports nothing wrong while acting on the wrong element, or produces subtly wrong data.
CauseTeach-a-task captures a visual path through a specific UI; the capture has no semantic anchor to survive a redesign.
DetectionOnly by re-testing against the live source. The most-repeated instance is a scraper that reported empty results hourly for weeks after a layout change.
FixRe-test after any website, connector or source-format change. Run the test twice with edge cases such as missing columns and empty records. Export run history into source systems, since only the 20 most recent run records per routine are retained.