Barbara Potocka and Józef Mniszech to Jakub Dunin, 1714–1716

01The documents and the evidence

All 29 source images were inspected, including backs, addresses and spreads. These are chiefly clear letters with brief cipher inserts; the tables below transcribe those inserts and the clear words just before them (the “clear anchor” column); the rest of the cleartext is not transcribed. Image numbers are DECODE image suffixes. Some letters have no year: placeholder catalogue dates have not been adopted as historical dates.

Potocka calls Dunin “mon fils,” but the university Women’s Court biography identifies her as his aunt. That source was found after the key recovery and supplied no cipher values.

02The recovered alphabet

The basic sequence runs backwards from 25=e through 14=x, and forwards from 26=a through 39=p, with a break before l. The Polish postscript uses 27 for a and distinguishes 31/32 in positions for l/ł; the exact diacritic rule remains uncertain. Attested higher homophones include 50=e, 52=a, 58=c, 68=n and 74=r. Doubling suggests values without fixing them: 14=x does not make 28=x, since 28=b.

Every letter value is inferred here (grade I), with doubtful readings marked M. None has grade H (original key) or C (independent known plaintext). The standalone figures 87, 94, 100 and 270 are unvalued person codes. A possible identification of 100 as the king is contextual only.

03R7524

Mniszech’s letter, 1714: none of its 232 numerical tokens is decrypted. The clear Polish and Latin words on the page give no decipherment. A tentative crib (hetmanow) and proposed nulls were tried and failed.

04R7525

Horyhorki and court influence: code 94 wants to be alone in the credit of 100; a house causes the writer harm; justice and enemies.

05R7526

The starosta is present; she asked whether he had arranged anything, and he has nothing; a journey to Lublin; a report that he was with the Dominicans at the time when Dunin danced with a woman in the field. The office/place name reads serecki and the reporting name Stabrowski in the numerical text; the letters of both are clear on the image, but neither person is securely identified. They are not emended to Merecki, serejski or Dąbrowski.

A cautiously word-divided reading, retaining the source’s irregular forms:

Pan starosta serecki iest tu. Pytałam się ieżeli ma co postanowionego. Ni ma nic. Iedzie ⟨dzie⟩ do Lublina. To tesz prawda że był u dominikanow wtenczas iakeś tańcował z nią w polu. Powiedał mi Stabrowski.

‘The starosta serecki is here. I asked whether he has anything settled. He has nothing. He is going to Lublin. It is also true that he was with the Dominicans at the time when you danced with her in the field. Stabrowski told me.’ ⟨dzie⟩ marks a syllable the writer enciphered twice (M): the figures run 21 25 30 16 21 25 30 16 21 25 30 35, iedziedziedo, evenly spaced with no cancellation, so the second dzie is taken as a slip of the pen; the alternative is the dialect form dzie for gdzie, ‘somewhere’. serecki is clear on the image (36 25 37 25 29 20 21, the 20 retouched in darker ink); the starostwo it names is not identified. Other irregularities are not silently repaired.

The division iakeś tańcował, ‘when you danced’, replaces iak estancował (corrected 30 September 2026; the letter values, grade I, are unchanged). It follows the writer’s own grouping: on the image the figures 21 27 20 25 36 (iakes) close a line as one group, and 19 27 34 29 35 17 27 32 (tancowal) open the next as one group, followed by a stop. A line end alone would not settle it, since the writer also breaks words across lines (udomini|kanow, powi|edal). But jakeś tańcował is ordinary Polish, the ending -eś of the second person placed on jak, while the earlier division needs a word estancował that Polish does not have, and a lodging verb such as stancować would need a second s that the figures do not give. The change makes Dunin, the addressee, the one who danced with her, and turns the clause into a date for his visit to the Dominicans; the earlier gloss, his staying with a woman, is withdrawn. On the image the first figure of the last group is 36 (s), not 39 (p) as first transcribed: its second digit closes at the foot like the 6 of starosta, while the 9 of 39 in polu on the same line ends in a tail. The transcription was corrected on 30 September 2026, and the name reads Stabrowski, an existing surname, instead of p. Tabrowski.

06R7527

The palatine of Vilna's obligation for sant mil fran on Bychow; measures to take and distrust of 94's help. The reverse concerns an accusation against the marshal of Lithuania, Biegański's confession to his father, a letter to 100, a letter through the grand chancellor, and an oath. The allegation is the correspondent's account, not an independently verified event.

07R7528

Rakiszki, the palatine of Mścisław, and property/reformation involving the first husband, Kroszyński. The second surname spelling is irregular (Krodyznski).

08R7529

Clear French says that since 10 May the clef is rompue and there is no way to loubrire. Another passage says someone has the clef in ceur. These report trouble using the cipher; they do not establish interception or cryptanalytic compromise. Later cipher words are rayne and maitresse.

09R7530

The grand treasurer, a confederation, the surrender of an army, and evidence concerning the arrival of a morcowide (probably Moscovite). Some number sequences yield corrupt spellings; the letter is not treated as a fluent verified transcript.

10R7534

Deputies, libertation, a doubtful zaia[73]d (possibly Polish zajazd), Bychowa, Białystok, Tykocin and Rybiński. Codes 87, 94 and 270 remain open.

11R7535

Children and a brother, being Saxon, Dunin's house, Chomiński and an ensign; court correspondence, the marshal and his father, Menzykw (possibly Menshikov), the queen and the prince of Courland.

12R7536

A disturbed seal and concern over correspondence; instructions to route letters via a name resembling Königsberg, followed by an unidentified Kinnik. The writer expects delay but greater safety. The reverse considers her security, lodging in case of war, keeping Dunin's wife company and a safeguard.

13Method and limits

Solver runs with a French language model failed. A simple-substitution solver with a Polish language model recovered the R7526 alphabet, and the key was tested on eight other letters before the remaining figures were assigned. Repeated words and names confirmed the higher homophones. Full-resolution crops and a final check of every page corrected transcribed figures, added the cipher on R7527’s reverse, and counted the standalone codes.

Mniszech’s different system was tried with Polish homophonic searches, a tentative crib, null and digraph hypotheses, and a frequency penalty. None produced coherent text. In a control, the same solver recovered only 49 positions of a synthetic homophonic ciphertext of the same length, 232 tokens. The failures on the letter therefore show nothing about whether it can be decrypted. It is left open.

Three Sanguszko key records (R7515, R7460, R7461) were inspected and ruled out by their signs or numerical values. Searches for editions, these correspondents and the shelfmarks found contextual references but no matching key or plaintext. Further correspondence is catalogued without online images located here. No earlier decipherment was found; that does not prove none exists.

14What remains uncertain

• R7524, all 232 cipher tokens: a separate system, no matching key or plaintext located, current solver also fails its control; not declared impossible.
• Potocka codes 87, 94, 100, 270, 23 occurrences: context suggests roles but does not identify exact entries; no matching nomenclator.
• R7534 exceptional figure 73 in zaia[73]d: one occurrence, no independent key, proposed zajazd does not justify assigning a value.
• M-grade names and anomalous sequences in reading.md, especially the unidentified starosta serecki and Stabrowski, the dittography ⟨dzie⟩, Krodyznski, connedere/morcowide, Keinisberk a Kinnik, tegn and tombap: image ambiguities or possible original enciphering/spelling errors, no parallel witness. Literal decryptions and editorial suggestions are kept apart.

Across all ten letters, 863/1,119 cipher tokens have a value (77.1%). The conservative coherence count is 745/1,119 (66.6%): it excludes entire doubtful segments, so it is a lower bound, not an error-rate estimate. Raw output, editorial interpretation and confidence are kept distinct in the complete segment reading.

15Sources and reproducibility

Archiwum Narodowe w Krakowie, Archiwum Sanguszków: the ten DECODE records linked in the document table. Related correspondence is catalogued as 29/637/0/1.3/9477, not established here as a duplicate of these letters. All ciphertext, the recovered key, the failed solver trials, the control and token-level measurements are in the repository. Run prepare_reading.py to reproduce the literal output and coverage.

Research notes, transcription, key and scripts. DECODE updates are queued, not submitted; the public reading files contain encrypted-passage extracts with gaps for omitted clear prose.