Letters I reported as specimens of a polyphonic cipher turned out to be more interesting than I thought. I learned this in Paolo Rosson's report of full decipherment of the letters (his github).
(Maxim Egorov worked on the same material and published his analysis, which came to my knowledge after writing the following.)
Back in 2017, I found some letters in BnF fr.3982, 3983, and 3984 employed a polyphonic cipher (see another article).
I noted two out of four letters from 22 July 1593 in BnF fr.3984 remain undeciphered:
(f.176) Baudouin-Desportes to Pope Clement VIII, Paris, 22 July 1593 (deciphered on a separate sheet)
(f.188) Baudouin-Desportes to Bishop of Lizieux, Paris, 22 July 1593 (deciphered on f.188)
(f.186) Baudouin-Desportes to Pietro Aldobrandini, Paris, 22 July 1593 (not deciphered)
(f.189) Baudouin-Desportes to Hieronimo Frachetta, Paris, 22 July 1593 (not deciphered)
(The name Beaudouin is taken from the BnF catalogue. It is also supported by the endorsement "Baudouyn Desportes" to the decipherment of the letter to the Pope (f.178v).)
Rosson noted these are part of a collection of intercepted letters and there is a record that these were not deciphered at the time. BnF fr.3983 has a list titled "Extraict de lettres interceptes tant en chiffre qu'aultrement."　Nos. 13 and 14 (f.141v) appear to correspond to the above letters (these letters are clearly endorsed "13" and "14"). It is recorded that they did not have time to decipher no.13 ("l'on n'a eu loisir encores de deschiffrer") and no.14 has not been deciphered, either ("qui n'a non plus encore esté deschiffrée").
On the other hand, the letter to the Pope from the same date (f.176 (endorsed "11" on f.179v)), accompanied with a decipherment, is also on the list of intercepted letters. It is wondered who deciphered the intercepted letters.
An obvious candidate is François Viète. While Cent Cents Colbert 33 does not record Viète's activity in the summer of 1593 (see another article), if this polyphonic cipher was also broken by Viète, it adds a new aspect to his talents.
A few hours after first posting this article, Rosson answered this by uploadeding a detailed report. The official decipherment was made by two clerks and the hand is different from Viète's. This report is worth further reading.
The date 22 July 1593 was immediately before Henry IV attended a mass to convert to Catholicism on 25 July 1593. The prospect is reported in a deciphered letter to Aldobrandini: "Et outre, M[onsieu]r le [Legat], sans consideration quelconque, veult que nous en venions a l'election d'un roy, lors que l'ennemy commence a gagner les voix de beaucoup qui luy estoient ennemis, par l'aparance qu'il fait de se vouloir faire catolique, et doit dimanche aller a la messe." (Moreover the Legate, with no consideration at all, wants us to proceed to the election of a king at the moment when the enemy begins to win the voices of many who were his enemies, by the show he makes of wishing to become Catholic; and he is to go to mass on Sunday (translation by Rosson).) This shows the reporter's irritation at the papal legate when Henry IV was about to gain popularity by conversion to Catholicism.
It may seem surprising that the letters were left undeciphered at this critical moment. Probably, decipherment of the letters to the Pope and the Bishop of Lizieux from the same date was considered sufficient. Now that full decipherment is provided for the first time, it would be interesting to see if these letters add something new.
In response to the above, Rosson compared the newly deciphered two letters with the other two deciphered at the time. Although the prospect of Henry IV's attending Mass was also reported in the letter to Lisieux, his model gave several points that are only in the new decipherment. As he says, we must wait historians' review but it is amazing AI can do this kind of job very quickly.
Rosson succeeded in deciphering the letters on 1 October 2026 with the assistance of Anthropic's Claude Opus 5.5. Even if the key is known, the deciphering is not straight forward with a polyphonic cipher. The solver has to try different candidates for every symbol to produce a coherent text.
An earlier attempt by Daniel Bourdeau had not been successful because automatic transcription was not accurate enough. (Still, it served as a starting point for Rosson's work.) Rosson's set of 6,794 labelled signs harvested from deciphered materials (larger than about 1,100 in the previous attempt) seems to have been the key to improve the transcription, hence the success.
As I originally observed, this particular cipher is tricky in that symbols for A/N, C/P, and D/Q all look like a variant of 4 and may be easily confused. It must have been troublesome for the contemporary decipherer, too.
The following is my partial verification of Rosson's decipherment. (I didn't try to look for other possibilities of the polyphonic assignment. I believe the coherent text of this length cannot be obtained by chance or tampering.) The reference to Mass is seen close to the end of this specimen.
(I made slight changes in the reading of cleartext. I substituted "en" for "icy" on the second line; deleted a question mark after one "de", and supplied "contredire", "bruit" on the 7th line.)
And the following is the image without decipherment. The reader may try to see how hard it is to decipher a polyphonic cipher even with a key.
