That’s an inference system issue… where SU recognizes and acknowledges that there are two edges next to one another…and it’s going to offer you the choice to snap to either one.
And I know you want the top side to be more heavily weighted in the snaps decision making here.
But I don’t think we are there just yet.
I think you’ll have to develop more sensitivity to how you approach competing snap points… and maybe if you’re not already doing it… refine some technique on how to lock inference directions via the Shift + Arrow keys.
The direction of approach can make a big difference too.
By Dropping in from the top, SU should be recognizing that top surface slightly before it starts tracking on the lower one.
That slim margin of lag time is the sensitivity I’m referring to.
There’s a delicate balance to be found between your normal modeling (e.g. drawing) speed, and the zoom distance you have in place.
Everyone has to learn to internalize this, before mastering the inference system… which typically is quite good most of the time.
Some nice comments can be seen in the following by @Geo, and others: