Exactly. Most of the article was based on some sort of unjustified "innate" knowledge of how users behave. He dismisses platform consistency as being unscientific and then goes on to present his own equally unscientific arguments for ignoring the convention. In the face of these two alternatives, I'd choose platform consistency every single time.

Designers work like this for good reason. Yes it reads like unproven pseudo-science, and it is, but there is reason behind this. When you work with thousands of concepts like this every day you can't empirically test them all across all cultures and environments (it would take forever and a day!) - you just develop a thought of what makes sense for you and have a basic think about why that might be so. This may not hold for all situations, but its going to be on average a better opinon than that of most other people (who will put a dialog box in one order or other without even stopping to think about why).

Kind of like how when you're going to choose a variable name, you don't do a linguistic study first?

Well, ultimately it's the balance between your time and you user's time.

Even if you don't actually do any empirical tests, a good user experience designer should at least IMPLEMENT their ideas and try them out first hand themselves before declaring them the best for a particular situation. That's why it's sad that so few user experience designers have any competency in actually implementing their designs. Many of them (the bad ones) don't even understand what the technology is capable of, let alone know how to program.

Ditto. Especially since the OS UI standards are usually based upon empirical studies conducted by MS or Apple.

Given an ideal test subject, that's what they'll do. Actual test subjects, on the other hand...

(paraphrased from "In theory there is no difference between theory and practice. In practice, there is")

That statement in the article is just wrong. A knowledgeable and experienced designer should be an experienced scientist who uses user testing to confirm rationalizations.

Yep. Testing 41 shades of blue gave Google products really nice designs.

Just because you're testing doesn't mean you're testing the right things.