Steps to reproduce:

1. Unzip the attached file to a URP project
2. Open "SG Precision Test/PrecisionTest Graph.shadergraph"
3. Inspect the different nodes configurations in the graph

Notes:

This test is made with a custom node that changes the output based on the resolved precision of the node. Red if "Single", blue if "Half".

The "refence" for testing can be seeing at the top, where only the node is used and preview is displayed : 

!image-2026-01-12-14-48-45-466.png!

Rows varies the input precision, and columns varies the node precision setting.

The second line of tests shows what happens when the node is inside a SubGraph, the subgraph uses a dropdown to switch the precision setting used on the inner node, and each group of 4x4 nodes use a different precision setting for the SubGraph instances : 
!image-2026-01-12-14-51-40-691.png!

Strangely, they all show the exact same result, independently of the precision setting of the SubGraph instances, or of the ShaderGraph global precision.
It is not clear why when the SubGraph is set to "Switchable" the instances still have individual precision settings if it has no influence and acts as if the SubGraph precision is set to "Half".

The last line of tests shows what happens when the SubGraph is set to a defined precision :
!image-2026-01-12-14-59-44-033.png!
These tests seem to be working as intended, the output matches to the precision set in the SubGraph asset. It is a bit unclear/uncertain of what should be the behavior when the node is set to "inherit" : it apparently uses the subgraph precision, ignoring the actual connected input that goes through the SubGraph connection. This is maybe expected.

All this exploration came from this issue raised on Discussions.