Spacedeck and stylus force guage

I am setting up a Nottingham Spacedeck with Spacearm I recently bought used. The instructions with the Spacedeck say to use scales directly on the mat. The instructions with my Shure SFG-2 stylus force guage say to place it on a record. This will result in a huge differance in pressure. Which is correct?
I've taken measurements with borrowed digital gauge between off-the-record and off-the-platter and they were different by .03g so the pressure difference isn't that huge after all.
Stylus force should only be measured in the plane of the record. On the mat and on a record are both too high. Try using playing cards to achieve the proper height.
The pressure difference will depend on the length of the balance and will be different with different scales. My ADC changes by about .15G from record surface to the plane of the record and a friends Technics changes by about .09G. These both have a much shorter balance than the Shure for example so it may very well change less.
I use a Shure with my Spacedeck. I no longer use the foam mat and have, instead, opted for a Mat1 from Boston Audio. As a coincidence, the point of thickness where the Mat1 and an average LP combine is within microns of the level where the Shure accepts the stylus if placed directly on the platter. So, I use the Shure on the bare platter and figure I am setting the force as it would be when the stylus ultimately touches the LP, which is what I think Marty is saying. I have increased and decreased the pressure from there and, right or wrong according to Not, this is where a measurement of VTF sounds best to me.
I agree with 4yanx's statement about getting it right by ear. Choose a method, find the VTF you like the sound of, and then continue to use the same method to check it or re-set it in the future.
Very succinct and practical, Twl. Cartridge designers and reviewers may need to know VTF exactly. Listeners just need to duplicate the optimal setting we proved by ear. As long as we measure consistently and repeatably, who cares if 'real' VTF is higher or lower than measured VTF.

I have a new scale on the way myself. Perhaps I can simplify the way I was planning to use it. Thanks, and thanks to Jsbail for asking.
Thank you to all who replied. Now that I know the answer, it seems so obvious!