That is why I clearly said it is very much dependent on the exact amount of noise! You can easily put to much in there, so that certain sequences will sound great, others really bad. I don't want the sample library to dictate what to write!
But on the other hand, trying to make everything very conform can also lead to a "dead sound". Maybe noise is not the perfect word do sum up what I mean ... just one example: VSL always recorded the open strings of each section with vibrato by scordatura. That is understandable by the idea of trying to make everything uniform and playable. But it is also one part of the puzzle of things you can do to steal the identity of the real string section ...
Another example: String players from time to time hit neighboring strings when playing ff ad above. Now, with VSL you will never find a note, where that is the case! I don't know it for a fact, but I could imagine VSL to muting the other strings with tape to achieve that goal!
There are occasions, where I might be annoyed by a Cello note having wrong pitches included (especially, it it is to much of that)! But in other circumstances it works much better to get a convincing performance.
From this dilemma, they could have learned! I am not saying they should have intentionally put as much noise in there as possible. They should have exactly calculated the amount of imperfections needed to make it convincing and created RRs for those imperfections to prevent any repetitive playing artificially making the imperfections stand out.
I have not produced a string sample library, so I am not claiming to know everything better, to be clear. I merely use those libraries and can therefore compare what helps certain libraries to work better then others. I can therefore imagine what could be done, but I might be wrong with the methods ...