OK, I fired up a Windows VM with ASCOM to look at the presets but I'm not sure what I'm seeing.
Offset should vary with gain. I checked two cameras, each of which has a preset for highest dynamic range, unity gain, and lowest read noise. They are:
ASI183MC-Pro
Gain 0, offset 10
Gain 111, offset 10
Gain 270, offset 10
ASI294MM-Pro
Gain 0, offset 30
Gain 120, offset 30
Gain 390, offset 30
These fixed offsets are contrary to what I understand about gain & offset. Then I started thinking this offset value was based on the maximum hardware gain, a common trick for standardizing your calibration library on a single offset -- but when I tested the ASI183 at Offset 10, it was pretty clearly clipping at gain 270. 20 seems decent, 30 seems safer still. The problem I need to correct is that when I originally took my dark calibration frames with KStars/Ekos, it used an offset of 8 (I must have missed correcting this). NINA used a default of 30, so they didn't match.
So, I don't know how these presets are supposed to work. Maybe I've got some ASCOM driver issue or something, though I did install the latest ZWO ASCOM drivers and it didn't make a difference.
I switched to the native driver for the ASI294MM-Pro and noticed that ASCOM default offset of 30 works great at gain 120 (unity), but is not enough for gain 390.
Then I thought I would try ZWO's ASIStudio software. That didn't expose offset at all, so initially I thought it must adjust it automatically... Gain "low" looked good. Gain "medium" seemed clipped. Gain "high" was badly clipped.
This seems like a missed opportunity for camera makers to deliver gain/offset lookup tables with their drivers (in ASCOM you can add your own presets, apparently). Software makers could provide a "calibrate offsets" button but this would be best just specified by the mfg. But that button could take some pictures to calculate a good offset for a given gain. For example, at Gain 120 it could take 2-second exposures at offsets in increments of 5 or 10 (starting at 5, never 0) until no pixels clip. This would be a more automated way to do what the human eye can do with a histogram. Then you could do this for multiple gains.
Another consideration would be to maybe make the default in KStars larger than 8; 30 seems safe, but I'm sure this varies a lot by camera.
Hopefully someone with more experience imaging and/or ASCOM can comment on this further.