Hunt for the key: two lost items from Kircher's correspondence and the one or two leaves that are missing

In September 2026 we obtained a number that overturned the project's plan: to recover the manuscript's key from a known text one needs about six hundred Latin words of aligned sample, and every known hint is two to three orders of magnitude smaller (see Anchor threshold). An awkward conclusion followed. If reading by statistics is impossible, what remains is to look for the key itself - the physical document.

It sounds like the plot of a novel. In practice it is archival work with a concrete list of holdings, and part of the list has already been covered.

A portrait of the artefact

First, what exactly to look for. Our model of the mechanism fixes the design of the key fairly precisely: six tables, each with three columns - for the start of a word, for the middle and for the end - over the 23 letters of the Latin alphabet. In all, about 414 lines of correspondences "letter - group of signs". Physically, one or two leaves with ruled columns. There may have been several copies: all five scribes of the manuscript worked from shared tables, and probably not in the same room.

Later we refined the portrait. The labels on the zodiac rings are two-letter abbreviations, and a quarter of them do not decompose by the text key; so beside the 414 lines there may have been a section for labels and diagrams. And the ring on leaf f57v - eighteen signs repeated four times - looks statistically like a reference record: the workshop kept working notes. The habit of writing things down is shown, not assumed; more in List pages.

The date of manufacture is 1404-1438 by the radiocarbon of the parchment. The region is northern Italy, Switzerland or Alsace, by the script and by the hand that added the months. Documents of this genre survive: the homophonic tables of Simeone de Crema (Mantua, 1401) lie in the State Archive of Mantua; the book of Francesco Tranchedino with 287 keys of the Sforza chancery (about 1475) is in Vienna, and it runs to 340 signs per system. By the standards of the period our key is of normal scale.

Two lost items

Now the trail. The manuscript surfaces in Prague in the 1630s with the alchemist Georg Baresch, who tries in vain to read it. In 1637 the Jesuit Theodor Moretus sends copies of the manuscript's script to Rome, to Athanasius Kircher - the best authority on scripts of his day. The letter itself is lost, but receipt is confirmed: Kircher replies on 12 March 1639, and that reply lies in the National Library of the Czech Republic. The copies have not been found.

In 1665 Johannes Marcus Marci sends Kircher the manuscript itself - with a letter kept today in the Beinecke under the shelfmark 408A - and with it "Baresch's attempts at translation". Those attempts are not in the Beinecke. If Baresch worked with some key material, it is there.

Two documented items, neither found. That is the niche: to our knowledge nobody has deliberately gone through the enclosures of the letters in the Kircher holdings for tables and alphabets.

Where the trail leads

Kircher's legacy followed a known route: the Collegium Romanum in Rome; after 1873 the holdings were hidden from confiscation; in 1912 part was sold to the Vatican - the very lot from which Wilfrid Voynich bought the manuscript at Villa Mondragone - and the rest stayed in the Jesuit archives, APUG and ARSI. Enclosures were often separated from their letters when the letters were bound into volumes. If Moretus's copies or Baresch's leaves exist, they most probably lie unrecognised in one of three holdings.

Kircher's correspondence - fourteen volumes, more than two thousand letters - has been digitised in full. The anchor units are known: Baresch's letter of 1639, two letters from Marci of 1640 and 1641, Kinner's letters of 1666-1667. The task is not the letters; they have been read. The task is the neighbouring leaves: tables, alphabets and pages of strange signs are recognised at a glance.

What has already been covered

Beinecke 408A-N. The inventory has been gone through by the box descriptions: from the seventeenth century there is only Marci's letter. No attempts by Baresch. Closed.

The verso of Baresch's letter. The only side of the 1639 bifolio without a scan. Two researchers who held the volume in their hands mention no signs on it; by a project decision we treat it as blank.

Hartlieb's miscellany of 1456. The prettiest thread. According to a catalogue of 1937 there existed a parchment miscellany of Johannes Hartlieb - a court physician who urged that recipes be hidden "in secret letters" - with eighteen secret alphabets on leaves 22-24, among them "Chaldean", "Egyptian", "Persian" and "from the Lithuanian land". Its whereabouts were considered unknown. We found it: the manuscript is digitised in Munich under the shelfmark Cgm 7958. The alphabets are foreign; the inventory of signs is not ours. But along the way it emerged that a ring of single-glyph cells, as on f57v, is a genre norm of German onomantic circles of the fifteenth century. What is anomalous is not the type of ring but its content.

Keys of the period. We compiled into one table all the open sources for ciphers of 1400-1460: the DECODE database (29 records in the window 1400-1440, syllabic signs none), the Ilardi reels at Yale (Gabbrielli's reconstructions from Florentine intercepts of 1424-1530; eleven keys of Siena), Meister 1906, three papers by Mark Knowles. The main find is the Milan cipher of 1424: a syllabary inside the radiocarbon window. It shifted our picture of the period, but no key gives the manuscript's profile: on Latin all of them give an entropy of 2.8-3.9 against 2.2 in the manuscript.

The Vatican. Ruysschaert's catalogue of the Collegium Romanum holdings (Vat. lat. 11414-11709, 716 pages) has not been gone through for anonymous tables and alphabets. It is known that researchers of the manuscript took only provenance from it. 181 items are not digitised. This is the next step, and it requires access to the printed catalogue.

What a find would mean

A document with the tables would turn the model into a reading tool within hours: the solver that is silent on the manuscript today reads 96.9 % of the text with a known key. Even a partial key - one table of six - would give an anchor above the threshold.

We do not promise that it exists. We claim less: two items relating to the manuscript demonstrably existed and have not been found; the genre of the document is known and survives; the route of the holdings is known; the niche is unoccupied. For a study in which everything else is closed by the numbers, this is the best of the remaining paths.

For understanding: a find would be recognised structurally - columns of Latin letters against groups of the manuscript's signs - not by reading. Any leaf that looks like a key will first be checked for a match between its sign inventory and the manuscript's, and only then applied.

← All findings