In my opinion, each scan must contain unexposed film borders. Such areas will make for true Dark point.
White point and neutral Gray is always a challenge if the scanned image does not contain such colors in it. But there must be a formula that allows to reliably calculate the proper values for each of these just based on the value of exposed area. It is done when printed optically and it is repeatable, hence there must be formula for each emulsion, exposure (under, normal, over) and development (push, normal, pull) combo.
From my countless experiments with scans of color negatives in PS, I have a feeling that RGB curves when used to set Dark and White points wreck highlights.
LAB might be a more suitable color space for the conversion. At least in PS.