I am not certain how/where it was collected (I am not the developer).
Also, the first word in the array won't be directly meaningful, because this is an array that represents a whole keyboard layout (so the first element is "a", the second "s", etc). The first real "word" is actually "and", which is not an interesting word. And it's first-ness is even less interesting because the order in this array is determined by the keys you press to select an item, not the frequency of that word ("a" has a low bit value, so "and" comes before "the", another uninteresting but very frequent word).
But, in later sections of the array (where you are pressing 5+ keys at a time to produce a longer word), you can choose from nearly any english word, and so it makes sense to choose ones that you use frequently, and so it makes sense to analyze a corpus of text to find the frequent words. The words ultimately chosen lead me to believe it was the developer's personal emails that were scanned.
Two of the test cases destructively modify the list, while the first does not. Of course they will be slower.