01The document, and the search for the ciphered originals
No. 97 is at folio 109 (Gallica canvas 227): “Coppie d’une lettre … de Charles de Lorraine à monsieur de Vaudemont”, Nancy, 18 June 1592. The volume is the duc de Nevers’s intercept and correspondence file for 1591–92. The task came with an instruction to check first for the ciphered originals of the letters intercepted in January 1592. They are not in the volume. No. 22, at folio 31, is a plaintext decipherment only: a fair copy of the French of three intercepted Vaudémont letters, with no cipher anywhere on the sheet, headed “Dechiffrement des lettres de Monsieur de Vaudemont à Monsieur de Lorraine [son] pere, interceptees et apportees à Monsieur de Dinteville, le XX[e] Janvier 1592”. DECODE agrees: its record for no. 22 (9444) is marked Decrypted. The other ciphered pieces of the volume (Birago nos. 31 and 36, Potier no. 79, Dinteville nos. 114 and 116) are Nevers’s own correspondents and their keys, not Lorraine’s. So no. 97 is a single copy with no sibling.
The letter mixes clear French and cipher inline on the same lines, and the postscript is clear. The clear text concerns the same business as the cipher: munitions promised and undelivered, Chaumont, the sieur de Buzonville, the attempt on Chasteauvillain.
02No key has ever been published
DECODE record 9449 is this letter, by shelfmark, sender, receiver and date, and its status field reads Non-decrypted; of the eight DECODE records for Français 3621 only this one and no. 116 have that status.
No key record in DECODE is attributed to Charles III or to Vaudémont. Keys for other Lorraine ciphers are published and are not this system: Bréval to Henri II of about 1620, and Bassompierre to
Charles III of 1 August 1593, both reconstructed by Satoshi Tomokiyo. Desenclos and Lasry’s 2024 paper on the Henri IV–Nevers ciphers of Français 3995 never mentions Français 3621, and Henri
Lepage’s 1864 edition of Charles III’s League correspondence stops in 1591 and prints no cipher table.
A prior reading did exist. The search above looked for a published key and missed a published claim. Satoshi Tomokiyo’s list of unsolved historical ciphers marks this letter (“Duke of Lorraine (1592)”, DECODE R9449) as solved by him in 2025, and quotes its opening: “J’ay este infiniment mal satisfait du peu d’execution …”. He published no key and no further text. The entry was already there on 6 September 2026, before this attempt; it was found on 24 September, after the reading, and was not used. The reading here opens with the same words, which independently confirms the first line.
Four 1592 key sheets in the Nevers key book, Français 3995, whose DECODE profile matched this cipher, were fetched and examined. None of the four is this key.
| Sheet | What it actually is |
|---|---|
| fol. 32 | A jargon dictionary of covert equivalents — Nouvelles = “Draps de soye”, Necessité = “Medecine” — in geometric signs and figures 2–40. Not a letter cipher. |
| fol. 83–84 | An all-digit cipher with a Nulles row. “Duc de Lorraine” appears in it as nomenclator entry 46, so it is somebody else’s key that mentions Lorraine. |
| fol. 98 | Endorsed in its own hand: “1592 / Chifre d’entre le Duc de Parme et le Roy Cath[olique] en l’an 1592 / lequel sert aussi p[our] le Duc de Sessa et don Diego de Ybarra”. |
| fol. 103 | A Nevers-circle nomenclator, its flap listing Gondi, Nemours, Guise and Montpensier, with three homophones for every letter — not a one-for-one pair cipher. It shows what a full key of this milieu holds: syllable figures 1–72, double-letter figures 73–98, word figures 99–353, and lists of places in which Lorraine is 30 and Metz 68. |
03The key recovery of 18 September (contributed)
An earlier attempt had treated the page as a cipher of invented symbols: segment the glyphs, cluster them by shape, solve the cluster sequence. The segmentation worked (the baselines slope, and a shear of −0.031 resolves the block into 21 line bands), but the clustering was unnecessary. The glyphs are ordinary cursive minuscules, the same letters as the clear text, some with dots or a bar, with a handful of special signs, and at one and a half times native size they can be transcribed directly from the page. The contributor’s transcription recorded 1,109 tokens: 1,071 letter symbols in 44 distinct forms, and 38 nomenclator figures. Its letter frequencies match French rank for rank through fourteen places, and its index of coincidence lies between those of a monoalphabetic and a fully homophonic system.
On a control with a known key of the same size, the random-move annealer of the earlier attempt recovered only 37–56% of letters at −2.64 per character while the true key scored −1.62: the failure was in the search. A steepest-ascent search with restarts from local optima recovers known keys at 98.7–99.6%. Two traps were closed on the way (free nulls that delete every hard position, and a degenerate etenete… key). The key it found was checked against the manuscript, not only the model: the eight-form group it deciphers as chasteau occurs twice in the cipher, and Chasteauvillain stands in the clear on the same page, in the body and in the postscript.
A 4-gram search over 44 free parameters can produce French-looking text from random input, so the pipeline was also run unchanged on shuffled nulls of the same symbols and frequencies, and scored blind:
| 4-gram score | Dictionary coverage | Distinct words of 6+ letters | |
|---|---|---|---|
| The manuscript | −2.10 | 83.0% | 33 |
| Five nulls, mean (range) | −2.73 | 74.6% | 1.2 (0–3) |
Thirty-three long words against none to three: there was French under the transcription. From controls with deliberately merged glyphs, the contributor estimated that about a quarter of the glyph
transcriptions were still wrong, and proposed re-transcribing the page with the key in hand. The 18 September decryption gave, for example,
chasteauilainss [145] nntaises deramenermonarme dooites esquartiers de-, with whole words correct and the joins between them wrong.
04The re-transcription of 22 September, and what it corrected
The re-transcription was done as proposed: the native IIIF image of canvas 227 (4046 × 5762), sheared −0.031 and cut into 21 line bands, each viewed in five crops at 2.3×, every word checked against the 18 September decryption. Laid out as pairs, the 18 September values already showed the system: glyph a gave i and glyph i gave a, o gave r and r gave o, q gave d and d gave q, n gave s and s gave n. Every letter of the hand stands for its partner in eleven fixed pairs, the whole 22-letter alphabet of the period:
| Pair | a / i | b / p | c / f | d / q | e / t | g / u | l / y | m / z | n / s | o / r | h / x |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Written a reads i, and so on both ways | ↔ | ↔ | ↔ | ↔ | ↔ | ↔ | ↔ | ↔ | ↔ | ↔ | ↔ |
| Sign | Value | Examples | Grade |
|---|---|---|---|
| two dots under a letter | the letter doubled: ṇ = ss, c̤ = ff, c̈ = tt | promesses, assister, passer, affin, remettre | H |
| looped æ | null, a word divider | peu · d’execution, belles · promesses | H |
| Θ | s | chemins, audits | H |
| 31 | ma (syllable) | com-31-nder, de-31-nde, 31 resolution | C |
| 139 | vous | chemins, 139 tiendrés; et 139 ne manquiés | C |
| 145 | me | affin que 141 ne 145 delaissent | C |
| λ, m̄, U, E, π, γ, ꜗ, low-barred t, figure-4 form | open | about 60 tokens | I |
| 13, 57, 88, 98, 103, 121, 122, 123, 137, 141, 146 | open | 17 tokens | I |
At least part of the nomenclator is therefore syllabic, like the syllable figures of the fr. 3995 keys, and not only a list of persons and places.
The 18 September key had 44 symbols instead of eleven pairs mainly because of two transcription merges. The transcription’s c was two glyphs, the secretary e (which stands for t) and c (which stands for f). The search had to send the merged form to t, so every f in the letter was lost (intiniment, satistait). This is why the old key had no value at all for b, f, g, x and z, a gap the earlier page had already flagged as the place where mis-transcribed glyphs had gone. Its B was likewise two glyphs, the beta-like b (p) and the looped secretary h (x, hence h): du peu but ceulx and d’execution. The dotted and barred variants that the search had spread over extra homophones are doublings. Correcting those turned most of the short connecting words of the old decryption into plain French with no further guessing.
What changed in the decipherment:
- prisonniers is pionniers: the Duke complains of pioneers and munitions promised, not of prisoners.
- 139 and 145 are not names. The earlier page guessed, from the clear text, that they might be Chasteauvillain and Chaulmont; in context they decode as vous and me (grade C).
- The f words come back: infiniment, satisfait, fait, affin, faire, fauldra, la Faulche.
- The chasteauilains slip that the null test exposed is confirmed at both places (L14, L18): the scribe left out the v twice, so it is an emendation, not a fault of the key.
- The measure: the old claim that about three quarters of the letters were right was an estimate from controls. Counted word by word against the new decipherment, the 18 September key and tokens give the same words for 449 of 1,105 tokens, 40.6%; the re-transcription gives 790, 71.5% (section 06).
05The decipherment
Running text of the re-transcription, line by line as the manuscript runs (lines numbered from L00; the unnumbered lines are in clear on the page). [?] is a word not deciphered; a word in brackets is doubtful
(grade M); bracketed figures are nomenclator figures not decoded. u/v and i/j are folded as the period writes them.
L00 [?] j'ay esté [?] infiniment [?] mal satisfait [?] du peu d'execution [?] vous
L01 [57] [?] et ceulx de la [98] [?] ont, de tant de belles promesses [141] [?] avoient fait,
L02 [l'?] de me assister et de [122] [123] [?] pionniers [?] et [?] munitions [?] [dressera]
L03 [?ien] fait. Vous [?] le faire l'entendre [?] clairement ausdits [57] [?], et les de-
L04 -putés de [?] [98], et vous ne [manquiés] à me [?] [demander] [?] [141] [?] conno-
L05 -issent le juste mescontentement [?]. Vous [?] dois avoir l'[137] [?]. Il [?] f-
L06 -auldra [?] de prudence, affin [que] [141] [?] ne me delaissent [?] tout à cou-
L07 -p [?] et de munitions et chevaulx [?] [121] [?] et aultres commoditez [?]
(clear) aussi si Dieu nous fault la grace de prendre Chasteauvillain
L08 [?] [57] me [?] demande [?] la plaine. I-
L09 -l fauldra [?] le remettre [13] ou le [?] luy [?] dire. Vous [?] escrire [?] afin de [?]
L10 recepvoir mon commandement [?]. Ne [?] mettre [?] quelq[u'un] qui, pour y comma-
L11 -nder et conserver la plaine [?] quant à ne vous, ceulx de [88] [?] [?] pour fa-
L12 -ire [?] passer [?] les [?] leur [?] ses promesses ne me [?] induiront [et re-]
L13 [-solument] [?ien] [?] le fera [?] le [?] des [?] ... apres l'effect [?] de
L14 Chasteau[v]ilains. Me [ne fais] [es] de ramener mon armée [droites] [?] quartiers de
L15 la Faulche [?] et me preadvertir [?] jour. Vous me [?] [pour] [?] me ma-
L16 -nde [?] chemins. Vous tiendrés [?] s'il ne vous [?] fault, pour tant faire paroistre ausd-
L17 -its de [88] ma resolution [146] [?ien] le peu de moien que [?] secours. Vous me promet[tés] par de-
L18 -là pour [?] le [?] Chasteau[v]ilain, ainsy vous l'on avoit [?] promis [?]
(clear) et de la ilz pourroient tascher de remedier, ou bien juger le peu de suiect qui me doibt occasionner
L20 [asseurté?] ses promesses [?] faire [?] passer oultre [?] [103] [?]In substance: the Duke has been infinitely ill satisfied with the little execution of what was promised. Certain people (57, “ceux de la 98”) had made so many fine promises to help him with pioneers and munitions; his son is to make it plain to them and to the deputies of 98, and not fail to demand of them that they recognise le juste mescontentement. Prudence will be needed so that 141 do not abandon me all at once, for munitions, horses and other supplies. After the clear line on taking Chasteauvillain, someone must be put there pour y commander et conserver la plaine; the Duke means de ramener mon armée ès quartiers de la Faulche (La Fauche lies between Neufchâteau and Chaumont, the area the clear postscript describes) and wants to be préadverti of the day. Vous tiendrés [les] chemins. He sets out ma resolution … le peu de moien que [j’ai] de secours, what l’on avoit promis about Chasteauvillain, and whether to passer oultre.
06How much is deciphered, measured
The rule, stated in measure_reread.py: the unit is the cipher token (one letter sign, a dotted double counting once; one special sign; one null; one figure). Each word of the decipherment is graded
H secure, C probable, M doubtful or I not deciphered, and a token counts as deciphered if its word is H or C. Every word is also decrypted mechanically with the pair key, so a decipherment that departs from the key shows up as an emendation.
| Tokens | Read | Share | |
|---|---|---|---|
| All cipher tokens (letters, specials, nulls, figures) | 1,105 | 790 | 71.5% |
| Letter signs only | 1,067 | 769 | 72.1% |
| Nomenclator figures (31, 139, 145 assigned) | 38 | 21 | 55.3% |
| Before: the 18 Sept key and tokens, same words | 1,105 | 449 | 40.6% |
Four emendations among the deciphered words: the two omitted v in Chasteau(v)ilain(s), q(u)e, and one promesses where the page has a special sign inside the word. At 71.5% the letter is below the 95% at which a text counts as complete, and the extent is recorded as partial.
| Figure | 139 | 145 | 31 | 57 | 141 | 88 | 98 | 13 | 103 | 121 | 122 | 123 | 137 | 146 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Times used | 10 | 7 | 4 | 3 | 3 | 2 | 2 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
| Value | vous (C) | me (C) | ma (C) | ? | ? | ? | ? | ? | ? | ? | ? | ? | ? | ? |
07What remains uncertain
315 tokens are not deciphered. Each open piece, with the reason it is open:
- Special signs with no settled value: λ (11 times), m̄ (9), U (8), γ (5), E (4), π (4), ꜗ, t with a low bar, the figure-4 form, and the reversed 3 in some places: about 60 tokens, and most of the words not deciphered contain one. m̄ and λ stand between whole words and may be nulls, but that is not proved. Grade I. Open because no key survives and each sign occurs one to eleven times, too few for context to fix most of them.
- Eleven figures: 57 (3), 141 (3), 88 (2), 98 (2), 13, 103, 121, 122, 123, 137, 146; 17 tokens, grade I. 141 is a plural subject (affin que 141 ne me delaissent), 57 and 98 are persons or bodies (ceulx de la 98, les deputés de … 98). One letter, no key, no sibling in the same cipher.
- Four garbled runs, about 45 tokens, grade I: L12 after passer; L13 between des and apres; L18 between pour and Chasteauvilain; L20 around m’a d…te. At IIIF 1:1 the glyph sequence does not decode to French under the pair key even with the specials set aside; the original or a higher-resolution scan is needed to tell e from c, b from h and the dotted forms apart there.
- Doubtful readings (grade M): dressera (L02), …ien (L03, L13, L17), manquiés and demander (L04), et re-/-solument (L12–13), ne fais, es and droites (L14), pour (L15), one vous (L16), asseurté (L20). Not counted as deciphered.
- Probable readings (grade C), counted as deciphered: the figures 31, 139 and 145; pionniers, belles, ont, avoient, l’entendre, ausdits, conno-issent, mescontentement, prudence, affin, delaissent, remettre, commandement, quelq[u’un], induiront, l’effect, vilain(s), préadvertir, jour, ainsy, l’on avoit, oultre and a few short words. Every other deciphered word is H.
- Not consulted: the Archives départementales de Meurthe-et-Moselle, where a duplicate or a key would most plausibly survive; the remaining key sheets of Français 3995 one by one. Both need physical access or a catalogue not online.
08Sources
- BnF Français 3621, no. 97, f. 109: Gallica btv1b52524472n, view 227; no. 22, f. 31, view 71.
- DECODE record 9449 (this letter, Non-decrypted) and record 9444 (no. 22, Decrypted).
- BnF Français 3995, ff. 32, 83–84, 98, 103 (the Nevers key book; eliminated).
- S. Tomokiyo, Cryptiana: the Lorraine keys (Bréval c. 1620; Bassompierre 1593). C. Desenclos and G. Lasry, HistoCrypt 2024, on the Henri IV–Nevers ciphers. H. Lepage, League correspondence of Charles III (1864).
- The 18 September key recovery: aryasn2026, pull request #4 (
ct/no97_reading.md,ct/no97_solution.txt). The 22 September re-transcription:ct/no97_reading2.md,ct/no97_reread.txt,measure_reread.py.
Manuscript rights: Bibliothèque nationale de France.
Transcriptions, key, scripts and controls: targets/lorraine1592/ in the repository.