That scale= is the worst thing i have seen in all pdks. Ever. I have also seen numbers like .scale=20 or even irrational numbers when shrinking a process... really evil. At the end there is for sure some calculation in area/perimeters where the scale factor is forgotten. Not even clear what happens in models, are geo parameters already processed with scale= ? or not... may be different simulators handle these things differently. This is a great source for errors.

That sums it up nicely. My understanding is that the scale is never applied inside a model, and only applies to specific dimension parameters (lengths, widths, areas, and perimeters) of specific devices. But as soon as there is a subcircuit wrapper around the device and there are parameters like "Weff = W - delta" it starts becoming very hard to pull the scalefactor out of all the equations.