Nice article Eric! I really like the example of a square that does not represent reliability due to the context and what's inside the square. You never disappoint!

The brain auto-completing the picture or missing elements is of course a big Gestalt principle -- might want to add a link or something for noobs :)

Also, I really like all your points, and I intuitively agree with things like "associations must not be primitive" -- but references would really help cement that and I do ask myself "is it really true that associations can never be primitive to be effective?".