x.com/CarterWChurch/status/2105167567706816690
1 correction found
Astra completed the entire multi-modal workflow in ~6 hours from a single image and goal.
The author’s own writeup says the process did not start from only one image: it also used a preexisting published key table and external French text statistics.
Full reasoning
In the detailed methodology the author says the work began from two files, not just one image: the scanned plate and a published partial table by Daniel Tant that already supplied 33 letter values. He also says those values were "taken as given" before the search began. Separately, he describes the solver as scoring candidate plaintext with French n-gram statistics drawn from Hugo, Dumas, and Marmont's memoirs. So the result was not produced from only "a single image and goal"; it relied on additional structured inputs and external language data.
1 source
- Breaking the Marmont Cipher, 1809 - Carter Church
The writeup states: "A partial table ... already documents 33 letter values... Those two files are where this work begins." It also says: "Tant's 33 letter values were taken as given" and the solver scored text against statistics drawn from "Hugo, Dumas and Marmont's own memoirs."