Cipher letters from European archives of 1503 to 1653, read or rebuilt in October 2026. Each page below shows one result in a form that you can check by hand: the key, a few lines of cipher, and the decoding token by token.
| Page | Document | What is new | State |
|---|---|---|---|
| oxford-1646 | An intercepted letter to King Charles I from besieged Oxford, 13 May 1646 (British Library, Add MS 72438, f. 10) | The first connected reading (an earlier partial reading by A. Aymeloglu had about 45 code values). The key is rebuilt from the printed Titus cipher of 1648, with a shift. | 89 % of the tokens read; the transcription is A. Aymeloglu's, and we did not see the manuscript |
| cocquet-1616 | Coquet to Mangot, Rome, 13 November 1616 (BnF, Clairambault 369, ff. 316-317) | First reading. Solved from context, then confirmed by the period key sheet (BnF fr. 18009, f. 156). | All cipher passages read |
| espagnol-318 | The viceroy of Sicily to King Ferdinand, Messina, 27 April 1503 (BnF, Espagnol 318, no. 94) | First reading. The key is rebuilt from the letter itself. | Read in large part; 64 code words have no value |
| dutch-1653 | The Dutch deputies in England to Boreel, 1 September 1653 (Thurloe, State Papers, i.435) | The key, and the link to the printed Dutch minute. The text itself is reproduced. | 127 of 131 groups fit |
| espagnol-132 | Three letters of Antonio Pérez, one page of Philip II, and a memorial of the Duke of Savoy (BnF, Espagnol 132) | First readings with keys that others published. | Five pieces read |
| gonzaga-1590 | Cardinal Scipione Gonzaga to the Duke of Nevers, 1586 to 1590 (BnF, fr. 4698, fr. 4702 and fr. 4696) | The cipher rebuilt with no key sheet; three period decipherments linked to their cipher letters; the rebuilt code tested on a second volume (97.6 % of 463 words right); letters and passages with no decipherment read, and checked against Pastor's history. | A code of 448 values; the new readings have gaps |
A seventh result, read for the greater part: centurione-1528, the "Venetian?" cipher with superscript digits of 1528 (BnF fr. 3022, no. 20, and the letters signed "Hieronimo Ranzo"). 86 % of the text reads as connected sense. The page shows the build of the cipher and worked lines. The key table, the read stretches, the blind test and the blind audit are in the folder centurione-1528. To check it by hand: the last line of the second "Ranzo" letter (BnF fr. 2988, Gallica view 20) has the groups m176, c193, v152, o66 with small numbers between them; they read "mille cinquecento vinti octo".
How the work was done
The work was done by a model (Claude, by Anthropic, running as Claude Code with subagents), directed by Paolo Rosson. No palaeographer or historian has checked it yet.
Most cipher lines had two transcription passes, the second one blind. Each reading has a control: a wrong-key or shuffled-key score, period glosses, or an outside source that was not used as a crib. Each page gives the numbers and the limits, and it says what is new, what is only reproduced from a period source, and what stays open.
Two earlier results
- pangoleen/desportes-1593: two letters of the Catholic League of 22 July 1593, read for the first time (BnF fr. 3984).
- pangoleen/senecey-1594: two League letters of 1594, read with the alphabet that the royal decipherers rebuilt (BnF, Cinq Cents de Colbert 33).
Five of these documents are on S. Tomokiyo's list of unsolved historical ciphers: the letter to Charles I, Cocquet's cipher, the Dutch ciphers of 1653, the Spanish letters of 1497 to 1504, and the cipher with superscript digits of 1528. All files are in the repository pangoleen/cipher-readings.