Comment by hooande
13 years ago
> A forger could have randomly changed Charles to "Charley", looked up an appropriate address, edited the metadata, come up with all the Hollywood touches, wrote the whole damn thing (quite an endeavour since relatively little material is borrowed from DN), and put it online.
Most people would have used that pile of circumstantial evidence to close the case. But you can't fool gwern (or L).
The best thing about this analysis is that it reflects the thinking of the one of the main characters in the anime. This seems like how the detective L would solve the same problem. has always been one of my favorite fictional characters because I greatly admire his (fictional) rigor and inductive abilities. Leave it to gwern to apply the same level of super detective skills to answering a question about the movie. It seems like a lot of work for a small point, but we'd all be better off if more people were this thorough in eliminating possibilities.
I saw the japanese movie and it was good, but probably best for people who were already fans of the anime. An americanized version might open up a much broader audience, though they'd have to put a lot of money behind it.
One thing I found strange about his analysis is that he kept on looking for new approaches after he got an upper boundary of 6.6% for the style result being random. I'd have introduced a more realistic model and looked at whether it get's me below 3% and be done with it (probably). He seems to like analysis just for the analysis sake. Of course it's still refreshing to see after reading tons of today's superficial blogosphere blabber (Google has ALWAYS blocked WP!!!11! ... wait no, sry).
I don't think you can consider the stylometric result as yielding such an upper boundary; yes, that's some strong evidence against one possibility and in favor of another, but we could easily find some more evidence which goes the other way - and indeed did.
So, it may be an upper bound _ceteris paribus_, but ceteris is not paribus here.