In the book Musical Human by Michael Spitzer, he comments how the score of a piece of music should explain how it should sound. He recognises that it doesn’t; the score describes what should be played.
What we fail to understand is the pace, cadence, expression, sharpness or how the notes are harmonised. What we have is an interpretation by every single person who reads the score – everything is the same yet different. If classic pieces of music that have existed for hundreds of years, ever changing, why should we assume that more simple instructions are interpreted EXACTLY as we wish them to be.
Herein lies the problem with trying to understand performance from a data perspective. There are so many variants that can influence the performance that to list everything one needs to consider would be unrealistic; if you’re going to do that, go the whole way and just automate it.
The smart learning people know this and work in a range of spaces and contexts, understanding the blend in a true sense won’t be an algorithm on its own.
Performance practice is always changing – performance practise is always changing.