A quick overview - in usability testing, testers are chosen, given an introduction to the software, and observed while they try to get stuff done with your product. This ideally involves two people aside from the tester - the "greeter", who interacts with the user/tester and guides them through the process, and an observer who focuses completely on watching the user at work.
Usually, two types of measures should be taken: numerical (number of errors, amount of time taken to find functionality, etc.), and qualitative (the user's attitude, comments - all the things you didn't think to measure in advance).
Quantitative measures are very important because then comparisons can be made - do users find functionality more easily/complete tasks faster with the "Edit" button hidden? By how much? How much longer do they take (if at all) to find that functionality when the button is hidden?
In sum, a usability study should be systematic - more like a psychology experiment or ethnography study than a product demo.
Pop-ups providing first use advice to users may improve usability results for new users. For experienced users, pop-ups are likely to decrease usability.
Placing many buttons on a screen will likely make it harder for new users to start using the interface. They will have to investigate many more options amongst the clutter, leading to more decision making ("what is this button? do I click it?") and a corresponding decrease in usability.
In the case of GitHub, the question of whether new and/or experienced users regularly need to use the "Edit" button needs to be investigated. The cost of making an interface control highly visible to new users may exceed the cost of experienced users needing to search for help on how to perform the rare action.
I was commenting on the question about usability testing (which yes, should include testing of experienced users); I don't have the information to make a judgment on the Edit button; I presume Google and Github both saw something in their testing that led them to make their design decisions. However, I am interested to know, in Google's case, what improvement, and what problems, they saw from this particular change. Anyone work on the new Google Contacts?
I appreciate that it isn't too much of a shibboleth for you and that you could stick around, I'd be interested in knowing a little more about the literature you mentioned. Do they propose specific techniques to usability testing? Which one do you use? In my case it's watching friends/family/expo attendants interact with the app but I don't have much of a formulated technique, it's casual, but I'd be interested in knowing more (and in the process I may be convinced that I should actually read that literature).
One absolutely can not depend on users to assume that needed functions can be found in invisible items.
I don't think it's that simple. It depends on the context and on the target. On a professional tool targeted to software developers like github, and being a consistent pattern, then yes, I think you can depend on that. If on the other hand you use that pattern in a context where it's not frequently seen, and/or it's not a consistent pattern in your app, you probably shouldn't depend on it, no.