There's a famous letter where Rasputin supposedly foretells his own death and the fall of the Romanovs — kill me by my own kind and the throne survives, kill me by the nobles and the dynasty is gone in two years. It reads like prophecy fulfilled. It's also almost certainly fake, and the tell is that it's too clean. Rasputin was a barely-literate peasant whose actual scrawl is a mess of phonetic spelling, dropped grammar, and no punctuation to speak of. The prophecy letter has the cadence and the commas of a literate hand writing for posterity. The forgery gave itself away not by saying something false but by being written too well for the man it claims to come from.
First framing: authenticity has a texture, and the texture is usually worse than a forger expects. The fake is betrayed by polish. A forger writing a peasant prophet reaches for the prophet and forgets the peasant — they smooth what should be rough because roughness feels like a mistake.
Second framing: this is stylometry doing detective work. Every writer leaves a fingerprint that isn't the content — the error rate, the punctuation habits, the sentence-length distribution, the spelling failures. You can't fake the fingerprint by trying harder, because trying harder is itself a deviation from the fingerprint. The harder you compose, the further you drift from someone who never composed.
Third framing: the unsettling generalization is that we authenticate against a prior of imperfection. We believe the rough thing because forgery is assumed to be effortful, and effort leaves gloss. Which means the better forgers get at faking roughness, the more this tell decays — and an AI that perfectly models a peasant's error distribution would pass exactly where the human forger failed. The defense was always the forger's laziness, not the truth's signature.
A concept page on internal-consistency authentication — the idea that a document is checked against itself (does the hand match the claimed author?) before it's checked against the world. Pairs naturally with the watch-vs-confession spark: one is about who made the record, this is about whether the record's texture matches its claimed maker. The essay angle: the forger's instinct to improve is the historian's best friend, until the machines learn to misspell on purpose.
Promote when it links to at least one non-Rasputin stylometric case (the Hitler Diaries, the Donation of Constantine's anachronistic Latin, a disputed authorship attribution) and when the "expiry under good generative models" claim is sharpened into a real question the page poses rather than a throwaway. Until the texture-prior is stated as a falsifiable method, it holds in LAB.