972Cyphersolver

Milan, Sforza chancery · homophonic substitution in graphic signs · 1450s–60s · attempted, not read

A Sforza chancery letter in graphic signs

A two-leaf cipher letter in the Milan archive's box of ciphers: the controls say Italian and a key was estimated from the signs alone, but the scans are too small for a reading.

Daniel Bourdeau · posted

Method: not solved · Extent: text not obtained

At a glanceNot solved · text not obtained

The document
A letter entirely in cipher from the Sforza chancery, no word division, undated; no sender or recipient is named.
Where
Archivio di Stato di Milano, Carteggio Visconteo-Sforzesco, Potenze sovrane, Cancelleria segreta, Chiavi e cifrari, cart. 1597, stamped leaves 158–159 (the pencil stamps suggest the leaves may belong to two letters, nos. 158/159 and 165; not yet checked); DECODE R5464.
The cipher
A homophonic substitution with about 60 graphic signs for the Italian alphabet. Dots separate look-alike signs.
How it was tried
A sign reader built for these scans, then a homophonic solver (Lasry's CTTS) with period Italian text, controls on shuffled and simulated ciphers, key refinement from words, and a hand correction of the sign boxes.
Why it resists
The DECODE images are 1,671 px wide and a sign is 15–25 px tall; about one sign in ten is misread, which is enough to stop the solver from locking onto sentences.
Still open
The whole reading. Better images from the archive are the next step; the key sheets tested so far do not fit this hand, and some in the box are 19th-century reconstructions.

01 The box and the letter

Cart. 1597 holds 174 DECODE records: 115 key sheets (nos. 1–114 and 169–171) and 59 ciphertexts. The key sheets are neat copies headed with correspondents' names (Cosimo de' Medici, Roberto de' Malatesti, Antonio da Pistoia, 1458–68) and the Naples inventory shows they are later reconstructions, so they are not a key to the letters. The letters use a dozen or so different sign alphabets, so every alphabet needs its own key; the catalogue lists them separately (items 2399–2413, one key per entry). This page covers one letter, R5464, which is long enough for a statistical attack: 3,279 signs, 60 distinct.

02 What the controls say

The solver's score on the real signs is compared with the score on the same signs shuffled. For Italian the ratio is 1.65; for Latin 1.38, Spanish 1.36, French 1.32, German lowest. Simulated Italian ciphers with 0, 8 and 15% sign noise were run the same way to see how much a misread sign costs. The letter families that come out are coherent by shape: the V-shaped signs fall on e, the φ-shaped on o, the hooked-stroke signs on n, the R-shaped on s.

03 The sign reader

The CTTS sign detector failed on these scans, so a reader was written here: deskew, straightened line strips, connected components merged by overlap, dots attached to the sign they belong to, then Ward clustering on shape features. Dictionary-word coverage of the decryption (5–12 letter words from a period corpus) measures it:

The last row is the answer to "can the boxes be fixed by eye": a little, and it stops paying off after about 30 lines. Resolution is the ceiling.

04 What does read

Short stretches build Italian words (dele letere, gravisima, stamatina tornare, molte cose, sempre desiderat amici, porta dela bona) but nothing builds a sentence, so by the project's standard the letter is not read. The sign transcription (3,279 signs, labelled s00–s59) is in the repository; the page images are not reproduced here because the DECODE copies are not ours to republish.

05 What remains uncertain

The next step is better images. A request to the Archivio di Stato di Milano for 400 dpi photographs of the stamped leaves 158, 159, 164 and 165 is drafted, not sent.

06 Sources

Notes, code and the sign transcription: targets/sforza1597/.

Cite this page

Daniel Bourdeau, “A Sforza chancery letter in graphic signs (cart. 1597, R5464)”, Cyphersolver, posted 7 November 2026, https://cyphersolver.com/sforza1597.html.