Gottfried van Swieten to Count Cobenzl, Bonn 1757

Summary. Gottfried van Swieten, later Mozart’s and Haydn’s patron in Vienna, began as a diplomat under Count Cobenzl, the Empress’s minister plenipotentiary in the Austrian Netherlands. From Bonn, the court of the Elector of Cologne, he reported to Cobenzl on the French armies in Germany during the Seven Years’ War, putting the sensitive sentences into numbers. DECODE holds three of these letters and records no decipherment for them. The undated one (R955) has the decipherment written under the numbers: the magistrates of Hanau debated neutrality in case the French left. It shows the system, an alphabetical list of letters and syllables with a few homophones. The two letters of January and February 1759 use a new table in the same style, with numbers 100–442. No key for it survives. It was rebuilt from the ciphertext alone by a solver (simulated annealing) over code-to-syllable assignments that keep alphabetical order, scored by a French language model. One random start in eight came out as French. The rest was corrected by hand against the cleartext around each passage. The letters report the convention that left the revenues of Prussian Gueldre to the French while keeping the Empress’s sovereignty, the principality of Moers handed to a prince without asking her, and a sharp judgement of Soubise’s army. That army lived on Gueldre’s contributions and planned no magazines, and among its generals only Broglie was competent. Extent: complete. About 21 of 553 groups in the 1759 letters stay open, nearly all proper names.

01 Three letters, two tables

Record	Folio, date	Cipher	Read
R955	f. 141, undated (1757–58, while the French held Hanau)	89 groups, dotted, 202–586	from the decipherment written below them
R957	ff. 327–328, Bonn, 14 January 1759	210 groups in two unbroken runs	all but four groups
R956	ff. 344–345, Bonn, 4 February 1759	343 groups in nineteen runs set inside clear sentences	all but about 16 groups

In 1759 the digits run on with no separators. They divide into threes, which the lengths of the runs confirm. Four DECODE transcriptions had dropped or doubled digits; those runs were transcribed again from the scans.

Page of numbers in groups of three separated by dots, followed by a French sentence in clear — R955, f. 141: the ciphered sentence, and under it the same sentence in clear. Image: Archives générales du Royaume, Brussels, Secrétairerie d’État et de Guerre inv. 1236 f. 141, via DECODE R955.

02 The 1757 table, from its own decipherment

The clear lines on f. 141 repeat the cipher word for word: “on a délibéré ici dans une députation de Magistrat sur le parti à prendre au cas que les François fussent obligés d’abandonner Hanau, on a osé opiner pour la neutralité et conclure même; on garde cependant un profond secret là-dessus comme de raison.” Aligning the two gives values that rise with the alphabet: a 202/203, ab 205, an 216, ar 220, as 223, at 224, au 228, be 241, c 250, d 279, da 280, de 282/290, en 308, er 313, et 315, fu 337, gi 343, ha 355, i 364, l 387, le 388, les 389, li 390, ma 396, ne 411, o 422, on 435, p 447, po 453, pr 455, que 465, r 469, ra 471, re 472, s 480, su 488, t 494, ta 496, ti 500, tr 503, u 511, ur 512. Each letter heads its own section and is followed by its syllables. The common units have a second number.

03 Rebuilding the 1759 table

The 1759 numbers do not fit the 1757 values, but if the new table was built the same way, it has a strong constraint: a higher number never stands for an alphabetically earlier syllable. The solver (msolve.py) keeps one inventory index per observed number, sorted, and allows at most two numbers per syllable. It anneals the boundaries against a French character model (fr-1750-nospace, added to the shared lang/ registry), with a small bonus per decoded letter. As a control, a 274-group French passage encoded with a random alphabetical table was recovered verbatim. On the real 491 clean groups, one start in eight gave “la nouvelle convention on laisse le revenu… en réservant la souveraineté… l’impératrice”, and the other seven did not.

The key was then corrected by hand, run by run, against the French sentence each run sits in. For example, “avant que les campagnes soient garnies, on ne songe pas à faire des magazins, si ce n’est aux dépens des États de l’Empire qui sont entièrement épuisés”. The end of the table falls into place: x 438/439, y 441, z 442. So do word entries filed in their alphabetical slot: nous, pour, qui, roi, comte.

Handwritten French lines with long runs of digits between the words — R956, f. 344: runs of unseparated digits inside the clear text (“le grand principe paroît être de faire 392 135 133 389 385 403…” = *subsister l’armée du roi y sans qu’il en coûte un sou*). Image: Archives générales du Royaume, Brussels, Secrétairerie d’État et de Guerre inv. 1236 f. 344, via DECODE R956.

04 Bonn, 14 January 1759

… de l’administration; la nouvelle convention laisse les revenus à la [G…] en réservant la souveraineté à l’Impératrice. On promet de toujours [?] de la part des François. Dans cette supposition, sans que nous ayons aucun ordre de la cour à ce sujet, et cela cause beaucoup de mésintelligences. On vient de la part de la [G…] de mettre en possession de la principauté de Meurs le prince de Con[…] […] qui vous formez des prétensions sans prévenir seulement l’Impératrice, qui en est souveraine.

The Prussian lands on the Lower Rhine (Gueldre, Moers) were in French and Austrian hands from 1757. The passage records a convention that left their revenues to one party and the sovereignty to Maria Theresa. It also reports a grant of Moers made without informing her.

05 Bonn, 4 February 1759

Il est constant que l’armée du Prince de Soubise est dans le meilleur état qu’on puisse imaginer … avant que les campagnes soient garnies, on ne songe pas à faire des magazins si ce n’est aux dépens des États de l’Empire qui sont entièrement épuisés; le grand principe paraît être de faire subsister l’armée du roi y sans qu’il en coûte un sou. La campagne passée en fait la preuve, aussi le prince [name] a-t-il avoué que les contributions de la Gueldre avoient suffi à l’entretien de son armée, et qu’on avoit même transmis assez d’argent à l’armée de Contades pour subvenir aux frais du traitement des troupes pendant l’hiver, et que toutes les opérations de l’année passée n’avoient eu d’autre but. Le prince pense bien, mais le reste est malintentionné; outre cela il n’y a pas un officier général entendu à cette armée, excepté le duc de Broglio; le major général (?) des Liégeois de Vaut est intéressé et ignorant, et non l’intendant nécessaire pour mettre en œuvre le système que j’ai détaillé ci-dessus … D’ailleurs, homme non violent, je n’ai pas besoin de parler de Dumesnil; il est universellement connu pour être prussien, […] le prince Camille … Parmi les maréchaux de ce[…] il n’y a que [name] qu’on estime. Comme de notre côté nous ne pouvons pas résister aux plaintes fondées des États, et que les ordres de la cour nous obligent de faire des représentations … on a pris occasion de former des plaintes contre le comte de Pergen comme s’il n’agissoit pas conformément aux intérêts de la cause commune.

06 What remains uncertain

232 (twice, R957): a name filed under G, the body that receives the revenues and acts on Moers. Not decrypted (I).
147 100 126 148 (R956): the prince who “avowed”. From the context it should be Soubise, but the numbers are out of alphabetical order, so this is probably a name group. Not decrypted (I).
170 323 441 (R957) and 170 271 289 337 (R956): the syllables decrypt (con…, …il l’on) but the words do not resolve. Conjectural (M).
290 313 316 313 373 (R956): the one marshal “qu’on estime”. The syllables decrypt to la-n-ne-n-r and the name is not identified; the digit division here is uncertain in the scan. Not decrypted (I).
“on promet de toujours” and “supposition [pr] sans que” (R957) are awkward French (C). “le major général” rests on a single number, 300 (C).

07 Sources

Archives générales du Royaume, Brussels, Secrétairerie d’État et de Guerre, inv. 1236; images and transcriptions on DECODE, records R955, R956, R957 (transcribers SofPe 2019, XL 2020).
No printed edition or earlier decipherment of these letters was found.

Transcriptions, solver, key and readings: swieten1757/ in the repository.

Gottfried van Swieten to Count Cobenzl, Bonn 1757–1759