Dear Hippolyte,
you've produced a fantastic and highly readable essay. The linkage of contextuality graphs to inconsistent cycles of propositions seems, in hindsight, almost obvious---but requires a great creative leap to realize. (Perhaps, to make this more vivid, you could've chosen a 'Penrose pentagon', the structure of which should essentially yield the KCBS-incompatibility graph. But I think the point is clear either way.)
Likewise, your introduction of levels of meta-description, paralleling the introduction of meta-languages in mathematical logic, is highly illuminating---ultimately, as 'paradox' in mathematical logic enters through the ability to formulate meta-language predicates within the object-language (for sufficiently expressive systems), it enters into physics with the application of the same theory to object and observer ('meta-object'). And ultimately, neither is, of course, paradoxical in the true sense: the Gödel sentence, unlike the liar, does not talk of truth, but of provability within some formal system---it is only through the misguided attempt of identifying this notion with truth that something genuinely paradoxical emerges. Likewise, quantum mechanics only appears to be paradoxical upon the---similarly misguided---attempt to assign truth values to all propositions, regardless of context.
My original idea for this contest was to try and apply my own ideas to the Frauchiger-Renner paradox, but I felt this would get to unwieldy. Hence, I'm more than happy to see it given such a capable discussion from your perspective. I hope this essay will do well in the contest!
Cheers
Jochen