It looks a bit wierd, cause its clearly what color ramp node are working, at lease partially…
Take a look:
Looks as expected. 0 instances on the left, max density at the right.
Our input which are taking vertex color are in the 0 to 1 range.
Same for color ramp node.
Same for distribution input.
But if we try to change our ramp it doesnt seems to work:
We can expect what spawning will start more from the middle of the plane, but everything looks the same.
Meanwhile, if we try to invert out values - everything works as expected:
So… what im missing?
Very good question 
Have you tried changing the ramp to simly linear? Ooor maybe adding a mapping node instead of the color ramp?
Looks like geometry should be more dense and subdivided. I guess color ramp in this case are actually dealing with vertices of geometry and if there not enought of them - results will look incorrect.
Well, then, did you make the plain itself with geo nodes, like a grid?
Or used a plain with subdiv and subsurf?
No matter what, try to increase the resolution (verticle-wise) and see if that becomes better
I am genuinely interested 
Yeah it was resolution of the plane. Previously it was just a quad.
More dense quad provides more expected results.
1 Like
Im still learnimg the new geo node settings… But it ses we cannot escape the need of geometry
I mean, they have to emmit from somewhere right? 