I'm pretty sure the testing methodology in this article is flawed. Using the manual switching codes (which are NOT recommended by Google) shouldn't be used to evaluate the vast majority of situations. For instance, doing timing on a manual network switch isn't necessarily equivalent to the phone automatically switching. Also, doing a manual switch and then extrapolating that the phone doesn't automatically switch back is not valid.

Basically, you can't test any of the automated functionality using manual codes!