diff --git a/src/optigraph.jl b/src/optigraph.jl
index 730351d..9e21ab0 100644
--- a/src/optigraph.jl
+++ b/src/optigraph.jl
@@ -503,7 +503,8 @@ the optinode variable value obtained by solving `graph` which contains said opti
 function JuMP.value(graph::OptiGraph, var::JuMP.VariableRef)
     node_pointer = JuMP.backend(var.model).result_location[graph.id]
     var_idx = node_pointer.node_to_optimizer_map[index(var)]
-    return MOI.get(backend(graph).optimizer, MOI.VariablePrimal(), var_idx)
+    # return MOI.get(backend(graph).optimizer, MOI.VariablePrimal(), var_idx)
+    return MOI.get(node_pointer, MOI.VariablePrimal(), var_idx)
 end
 
 """
diff --git a/src/optimizer_interface.jl b/src/optimizer_interface.jl
index 8b9c39f..f9946d7 100644
--- a/src/optimizer_interface.jl
+++ b/src/optimizer_interface.jl
@@ -362,6 +362,9 @@ function JuMP.set_optimizer(node::OptiNode, optimizer_constructor)
     return nothing
 end
 
+# NOTE: this resets NLP data on every NodePointer, so graph solutions get cleared
+# This is currently a known limitation in Plasmo.jl. If you solve a node after a graph,
+# it will remove the graph solution.
 function JuMP.optimize!(node::OptiNode; kwargs...)
     JuMP.optimize!(jump_model(node); kwargs...)
     return nothing