Heh, heh, that looks familiar. As I said at the time:
Another method for gradient points (culling) that makes more sense to me, and I’ve used it several times, is this one from January, 2024:
gradient_points_2024Jan17a.gh (28.6 KB)
It is based on Random Reduce in a range of “consecutive domains”.
I looked at the files in this thread but the GH file doesn’t seem to be connected to the hidden layer in the Rhino file. Did you make the connection yourself?
I can do that but… 
OK, I’ll stop whining now. This is a simplified variant of the idea:
gradient_2024Oct18a.gh (172.4 KB)
Geometry is internalized, no need for Rhino file.
P.S. The white group in version ‘b’ below sorts the branches by Z.
gradient_2024Oct18b.gh (174.7 KB)
Oops, I see an error… 