if you ignore the slurs and pretend the piece is written in 4/4 (which I do for the sake of this), the last measure actually could plausibly precede the first measure (this happens where the sheet music shows the direction "du bout de la pensée"). But I don't think it would be as interesting a listen, and it wouldn't have been a very fun project.
Well, it depends on how you feel about "appropriating" the original vs. feeding it to a Markov chain. I don't think less code/no code is necessarily boring, nor would a more complex LSTM model of the music necessarily be more interesting.
I was thinking of a non-digital piece by Rodney Graham, where he retypeset a few pages of a book so that a looped reading became possible:
"In Lenz of 1983, for instance, Graham appropriated an unfinished novella by Georg Büchner in which the mentally unstable protagonist, Lenz, wanders through a forest. The artist reset the type on pages one to two and on pages five to six of an English translation of the story, to set up a situation in which the page breaks at the same phrase ‘through/the forest’, which establishes a loop whose effect is to cause Lenz to endlessly repeat his journey by returning from page five to page two; in 1993 (fig.6), Graham had a reading machine constructed for this version of the text to facilitate the repetition, and the hardcover version of the 1983 work was a book bound with eighty-three repetitions of these pages, which in principle can be continued forever."
