It appears that LR performs simple piecewise interpolation of a slider value between the slider values of the images (ISOs) that you feed it to construct the preset. It is logarithmic, like ISO, but appears "linear" when you plot a slider value as a function of log(ISO). See attached.
To probe the behavior of LR's new preset thing, I shot a sequence of images at 1/3EV ISO increments. I made two presets: one where I specified the luminance NR slider setting at base ISO (160 for the Fujifilm X100 V) and ISO 6400; and one where I specified the NR slider setting at ISO 160, 200, 400, 800, 1600, 3200, 6400 and 12800. I made presets of each of these conditions and then applied the preset to the sequence of images and read off the NR slider values that LR chose to fill in the blanks, so to speak. In the plot, the NR slider values with points were specified at their respective ISO values - the line(s) between points represent LR's interpolation. The more ISO values (with specified slider settings) you feed LR, the more control you have over shaping the behavior of the preset. Also, it appears that LR does not perform extrapolation - so if you feed it a base ISO and an ISO 6400 image with various slider settings, for example, anything higher than ISO 6400 gets the same settings as the ISO 6400 settings.
kirk