The way VSA worked was easy for anyone using it to understand. This was because VSA generated an intermediate plain language file that was then compiled and processed.
Want to see how it worked? Read the source code that was fed to its compiler.
This situation gave a huge advantage in that any program that could be gotten to talk could be included into the VSA analysis - Have a cylinder combustion chamber temperature program that depends on the compression ratio? Take the outputs of the varied diameter, stroke, wrist-to-crown variations, et al, and use those to generate temp outputs and sensitivity contributors to same.
The feature VSA did not have was the ability to handle variation due to default angular plus/minus tolerances. It did best with features controlled by feature control frames and their geometric characteristic symbols. OTOH nobody I know handles default angular plus minus tolerances in their analysis.
Vector loop or Monte Carlo, shop floor info is only necessary if yields are to be predicted, otherwise it doesn't matter. At least the models I built never depended on shop floor data. We just checked to see if, when within tolerance, the parts would function.