(Part 1) (Part 2)
In any case, the genre-defying twist is in many ways invited by the very conventions of the genre, as if distortion of the form were built into the form itself, and some of the most 'unconventional' solutions are actually found in the most classic and even genre-defining stories. Even discounting ‘unusual’ solutions that are nevertheless perfectly permissible in the ‘classic’ canon (suicide disguised as murder, etc), the genre has from the get-go been corrupted by every manner of perversion. Unclassical solutions in ‘classic’ stories (Christie, Dickson Carr, Chesterton, Poe, Collins) include the following: the narrator did it; the victim did it (not suicide); everyone did it; no-one did it (there was no crime); a non-human animal did it; the criminal doesn't know he did it (but it wasn’t an accident); etc.
The Oulipopo (note that reduplicated syllable: Ouvroir de Littérature Policière Potentielle) was founded in 1973 as a sub-commission of the College of 'Pataphysics. François Le Lionnais had in 1971 written the founding text of the ‘analytical’ wing of the movement (the other being the ‘synthetic’, concerned with the composition of detective stories under various constraints), a study entitled ‘Who is Guilty?’, in which he attempted to outline all the possible solutions to a murder mystery narrative. (My information here all comes from the excellent Oulipo compendium, ed. Harry Matthews.)
Le Lionnais, who had clearly read many more crime stories than I have, even knew of examples of the following: the detective did it; the author (who is not the narrator) did it; the publisher did it. There are even apparently instances of the most genre-defying and perverse solution of all, that is, the absence of any solution (‘we can never know’) which ramifies, in Le Lionnais’s scheme, into three distinct types.
It is one of the most widely-acknowledged rules of the genre that the detective cannot be the culprit (Van Dine reserved particular scorn for this gimmick). I don’t know which story Le Lionnais had in mind here; but I know of one interesting example of the detective-as-culprit: G. K. Chesterton’s ‘The Secret of Father Brown’ (‘You see, it was I who killed all those people’) – where the guilt, however, is moral and spiritual rather than legal.
Clearly some of the instances mentioned here are not properly speaking canonical representatives of the genre, but parodies (for example, the one in which the criminal is the publisher is a ‘humorous story by P. G. Wodehouse’). Still, it is clear that the mystery genre has always clamoured to violate its own generic conventions. Like all genres in fact.
The only twist that had never been done, to Le Lionnais's knowledge was: the reader did it. The Oulipo Compendium duly supplies an example of just that, in Jean-Louis Bailly’s La Dispersion des cendres (1990). (Although it has to be said that this cheats a bit by being merely a description of such a novel and not the novel itself; and by really being a case of ‘the purchaser of the book did it’ rather than the reader.) There are other examples, the internet tells me (pdf), and I’m disappointed – if unsurprised – to see that I’m not the first to think such stories should be categorized as ‘youdunnits’. It appears that the examples mentioned in that link adopt the expedient of the second person narrator. That may seem like a cheap trick, reminiscent of those ‘choose your own adventure’ books we used to read as children, but Michel Butor’s 1957 novel La modification (although not itself a murder mystery) proves that it can be done well.
Incidentally, I have arranged it so that the email notification for one randomly selected comment on this post will remotely trigger a mechanism causing a hammer to smash a flask of poisonous gas concealed in my enemy’s bedroom. You have been warned…