One major role of an aggregate data reports writer (i.e. "Data Scientist") was Business Analyst. You can do all the tests you want, but when the actual users are deliberately using piece of data A, which means B, and if you have documentation at all is listed in the documentation as meaning B, for Q and T (depending) instead, it's entirely possible that your tests will show data A as being about as valid as your other data... and without knowing it's actually sometimes Q and sometimes T, using A to show B generates incorrect results.