I plugged my Landscape data and the actor data (of the actor where the PCG component is), tagging a static mesh inside it. The difference is showing in its output that $Density is there, but the Attribute Noise is not being able to see it. Why?
attribute noise is not finding $density because of how the union of those two sources lands. the landscape branch produces points that carry $density (surface sampling writes it), but the actor-data branch — points derived from a static mesh — has no density attribute at all until something creates one. when you union heterogeneous attribute sets, pcg keeps the union of attributes but points missing one simply have nothing there, and attribute noise errors on the points where the selected attribute does not exist. so the noise node is not broken — half your input literally has no $density to read.
fixes in order of preference: (1) move the attribute noise onto the landscape branch BEFORE the union, where $density is guaranteed to exist, then union after; (2) or give the actor-data branch a density first — a density node (or a surface sampler over the mesh) writes $density, so the union ends up with a consistent set; (3) if the goal is noise-driven culling, remember $density is what density filters consume — either run attribute noise in density mode with min/max inside 0-1, or write to a custom attribute and gate with an attribute filter instead.
also double-check the input pin/attribute selector on the noise node — with a union in the chain it is easy to be reading the wrong tag. but the root cause here is the missing attribute on one branch, not the noise settings.