I see you point, but I saw this "bed roughness' more like: there are a lot of hills on the plater: so a positive value , or: there are a lot of potholes in the plater, so a negative value to define them.
"Hills" and "Potholes" are relative to an arbitrary datum. The relationship between them stays the same, only the start point of measurement is different. If the hardware Z-level is set to 0 when the nozzle contacts the highest point, all measurements from that point are lower, which is the only way it can be set--otherwise you drive the nozzle into the bed elsewhere. Use KS Z-offset to fine tune the actual gap adjustment, if any, as hardware Z-level (at least on this machine) is a coarse setting compared to what I can obtain with KS Z-offset.
So, Bed Roughness can only be a positive value. It makes the first layer taller and computes the additional extrusion volume needed to compensate for the additional cross section that it needs to fill.
What has always puzzled me is that it really doesn't work as you would think. In order to take care of that which are always "potholes" one must compensate for the added layer elevation by dropping the nozzle back, at least partially, to where it would be if Bed Roughness were at 0 by setting a negative Z-offset that is a percentage of the Bed Roughness value. I don't recommend completely offsetting Bed Roughness additional elevation with an equal negative setting in Z-offset. The bed is rough, so there will be areas that haven't been identified as being at or below the point used to set the hardware Z-level. If I do remove some of the additional elevation, it is only by 25%-50% of the Bed Roughness value in the negative. This makes an increased "squish" in the higher areas, leaves some tolerance for areas that are higher than unexpected, and fills the lower areas; creating a flat plane upon which the remainder of the print will build.
I like that there is the separation of controls with which to fine tune this but new users would be potentially confused by Bed Roughness not making the "vertical autocentering" computation on its own. Perhaps this functionality could be added into the Beginner Settings Level.