Hmm, frequency analysis didn't work well in this cipher, so I decided to do it systematically ... I created genetic algorithm expecting there are dummy and deletes as in C expecting just each letter is used just once.
I got first 10 characters quickly, than I got almost readable rest ... I of course should use results of genetics just as good starting point.
(I used frequency, digraphs and about 150 most common words with guessed weights as evaluator.)
I was surprised there were no new features, just quantity what made this cipher more complicated.
OK, there was probably new feature ... E->AN. I didn't consulted the original where Rowling uses a bit deformed English often...