Additionally we could set up some template scenes for examples. So we know the node preview window is good for illustration and we have spheres being used for some texturing and displacement shaders. A locked off camera on a sphere would help get some consistency between examples and some basic rules on naming shaders. So there is a consistent language being used between contributors.
For example a Constant Scaler of 10 is named "10".
A Constant Colour of RGB 0.5,0.5,0.5 is named "RGB 0.5,0.5,0.5"
A Constant Vector is named "V 100,1000,100"
Which will show static values being used without having to jump into the node to find it, as WAS has shown in his post above.
Notes Nodes should be used where ever relevant to describe what is happening at that branch of the network. Even if it's just a label to help your written explanation.
Nodes should be laid out tidily, with a clear view of the connections and broken into well spaced groups to separate processes. For example converting a scaler input into radians as one section, feeding a Sine function another, isolating the positive part of the curve another. So the flow is broken into easily digestible blocks that people can follow to learn and or problem solve a contribution.
These are just a few ideas.