Real world testing yields messy, less-than consistent and somewhat subjective data, but it is describing the EXACT thing we are most interested in.
Roller Drum tests yield precise, quantifiable, reproducible, and consistent data.... but.... its relationship to what we really care about is unclear..
The former is an example of measuring what you value.
The latter is an example of valuing what you can measure.
Unfortunately people are too often seduced by the latter. They are essentially mistaking precision for accuracy.