01 The letter
The letter is dated “Dall’Isola di S. Principe, 8 8bre 1637” and signed F. Guglielmo Vizani. The address leaf, f. 293v, reads All’Ill.mo … Sig. Conte di Cesy in Constantinopoli, so the recipient is the French ambassador, Philippe de Harlay, comte de Césy. It is in the same volume as a second letter of the same date (f. 294), which has one cipher line. The text is clear Italian with cipher in the lines that matter; eight lines are enciphered, 54 words and 190 signs. The clear parts say that on 30 May the writer’s orders were changed, and that he will say everything once the plague has passed, since he does not want to trust it to the pen.
02 The reading
As read by Satoru (name codes in brackets, unread):
Il signor [231] m’ha mandato a domandare se li quattro mila scuti […] a [31] di monsignor Beria d’ordine di V.E. furono promessi a detto signore, … che è vicino all’ottenere il patiarchato per sé. … [64?] Monsignor Beria spera che tra due mesi sia per esser mutato; il [36] e dattone(?) un suo partiale, e per sua [198] havere l’intento a [227] è incertissimo.
Signor [231] sent to ask whether the four thousand scudi […] to [31] of Monsignor Beria were promised, by Your Excellency’s order, to the said lord, who is close to obtaining the Patriarchate for himself. … Monsignor Beria hopes that within two months [the Patriarch] will be replaced; [36] … a partisan of his; and whether, through his [198], he can achieve his aim at [227] is most uncertain.
Satoru identifies Beria as Cyril Kontaris of Veroia, backed by Rome and later Patriarch Cyril II. That identification is his; it fits the history he cites: Cyril Lucaris had regained the Patriarchate in March 1637, Kontaris took office on 20 June 1638, and Lucaris was strangled a week later. His reading of the cipher line on f. 294 (“il Veria … sia lontano”, with the same key) was not checked here.
03 The cipher
A homophonic symbol cipher. Words are divided by a stroke. The vowels have two or three signs, most consonants one or two. There are signs for tt and ss, word signs for il and che, and numeric name codes. Letters that differ only by a dot or a stroke on the original were separated by Satoru on the enlarged image (Ω̇ = i against Ω = g, ∸ = o against ÷ = h, + = c against † = l).
| letter | signs | letter | signs | letter | signs |
|---|---|---|---|---|---|
| a | ◇̇ ⌒̇ 6 | l | † ‡ | s | ⊢ ⟷ |
| b | □ | m | ꞗ | t | π |
| c | + | n | ȣ đ | u / v | υ |
| d | ɔ | o | ∸ # | tt | ꝋ |
| e | ʂ ẻ | p | ʋ | ss | Ƶ |
| g | Ω | q | h | il | Φ |
| h | ÷ | r | ⊣ | che | ⌂ |
| i | Ω̇ 8 |
Glyph shapes are approximated with look-alike characters, as on Satoru’s page. Name codes written as numbers: 31, 36, 64?, 198, 227, 231.
04 How it was read, and what was checked
How. Satoru cut the eight cipher lines from the Gallica image (8674×6462 px), transcribed them sign by sign, and had Claude recover the key from the ciphertext alone by simulated annealing scored against Italian four-grams with word boundaries (a model built from I promessi sposi) and a word list. Five of eight random seeds converged on the same reading. No existing key or decipherment was used.
Checks here, none of which repeats his search:
- Re-decoding. His key applied to his 54 transcribed words (
dec.py) gives the plain he lists for 46 of 48 words. The other two, both monsignor, lose an s and an n in his glyph strings (an abbreviation, or a slip in the transcription); the rest are name codes. - A control. The decode scores −1.78 per letter under a model of sixteenth-century Italian, against −4.30 for the best of 200 keys of the same shape with the letters shuffled among the sign classes (mean −5.30). Under a modern Italian model the figures are −1.99 against −6.66.
- The image. The first nine words of line 1 (il signor [231] m’ha mandato …) agree with the Gallica image sign by sign, and the code 64 before the second Monsignor Beria is two digits, as he has it. The rest was not compared by eye.
- His own controls, taken from his page and not re-run: 16 synthetic ciphers of the same size were solved in 15 cases (97–100% of letters); 24 shuffles of the text scored −1294 to −1236 against −1073.7 for the real one.
05 What remains open
- The six name codes: 31, 36, 64?, 198, 227, 231. Candidates are the Patriarch, an Ottoman official, the Mufti; none is confirmed. Césy’s minutes for 1637–38 (BnF fr. 16154–16155) are microfilm only, and no Vizani or Césy code list is known.
- Two unread signs (the word after se and the one after scuti) and the queried dattone.
- An independent transcription from the image. This page rests on Satoru’s; two glyph strings in it are a letter short, and his page says 53 words where it lists 54.
06 Sources
- Satoru, The Vizani Cipher (1637) Deciphered, 1 October 2026: transcription, key, concordance, controls. Made with Claude (Opus 5.5) under his direction. Also sent by email.
- Bibliothèque nationale de France, Français 16158 ff. 292–294, Gallica.
- Cryptiana, list of unsolved historical ciphers, where the letter is listed.
Notes, his transcription and key, and the decoder with its control: targets/vizani1637/.