Yes, we can measure and obtain results that will no doubt get misinterpreted and/or overgeneraluzed. In time, we might even sophisticate things and have robot execute different actual strokes with sound consistency and near absolute reproducability. And measure velocity, exit angle, rpm.
We’re not there yet. Even then, variations in technique applied will produce varying results. A suitably wide array of strokes to cover might tell enough.
The thing is, once you start measuring you need to beware. Your tests implicitly assume a reality model, and that is often poorly understood by those interested in using test results as performance indicators.