What the study is built on, and how to reproduce it
Data & code
Updated 31.07.2026
Every number in both papers is produced by a script with a fixed random seed, and each maps to a specific line of a specific file. Nothing here rests on a figure that cannot be regenerated. This page lists the sources, the corpora, and the analysis code.
Primary sources (public)
The manuscript itself and its transliterations are open data hosted by third parties. We use them unchanged and link to the originals.
IVTFF 2.0 text
Main line: ZL3b-n (Zandbergen-Landini), 5389 loci, 37,712 tokens.
Robustness on IT2a-n (Takahashi) and others.
213 page images
Full-resolution photographs of Beinecke MS 408, indexed by section and folio.
archive.org/voynich →Yale IIIF
The library's canonical folio order (213 canvases), used to map every scan to its folio.
Yale Beinecke →Corpora we assembled
The comparative material - cipher reconstruction, plaintext controls, and the degree-name candidate lists tested in the negative program - was normalized into machine-readable form for this study.
| Corpus | Role | Origin |
|---|---|---|
| Naibbe tables & deck weights | Cipher reconstruction (Paper I) | Greshko 2025 (doi) |
| Cato, Pliny, Antidotarium Nicolai | Latin plaintext controls (genre gradient) | Public editions, normalized |
| Usuarium liturgical calendars | Saints'-calendar candidates | Usuarium, ELTE |
| Lleida consueta RC.0031 (1400-1457) | Anachronism control (Paper II) | Fité i Llevot 2015 |
| Astrolabium planum 1488 - 360/360 degrees | Degree-image catalogue | BNC Firenze, transcribed here |
| Alfonsine star catalogue (V/61); cisiojanus; Liber Hermetis | Further crib candidates | VizieR; Grotefend; Gundel 1936 |
Analysis code
100 Python scripts, each self-contained, each with a fixed seed, each writing its result to a same-named report file. They fall into four groups:
- Parser & base statistics - one shared IVTFF parser and five statistics modules (Zipf, entropy, robustness), validated against published figures.
- Mechanism (Paper I) - the generative model and its tests: statics, the homophone-twin layer, seriality, the four boundary signatures, the inversion controls, and the genre gradient.
- Labels (Paper II) - the four sigla tests, and the three-stage crib pipeline with its seven rejected source classes.
- Key-recovery attempts - the external partition test and the internal hill-climbing attack, both of which returned honest nulls: the key is not recoverable by either route.
A registry maps every value in the papers to its script and line, so a claim can be traced to the exact computation that produced it.
Reproducibility rules
Fixed seeds
Every simulation is deterministic; re-running a script reproduces the same number.
Pre-registration
Success criteria are written into the script before the data are examined - no test is re-aimed after the result.
Number ledger
No value enters a paper without a ledger line linking it to a script and marking it verified.
No readings
The code measures structure; it never outputs a "translation". A coincidence is not a decipherment.
Availability
The complete replication package - 208 script and report files, the normalized corpora, the
zodiac-labels dataset (270 ring labels), the claim registries for both papers, and a
run_all.sh smoke test - is available for direct download:
voynich_replication_v1.zip (3.3 MB) ↓
Both preprints are archived on Zenodo; each record carries the same replication package and a citable DOI:
- Paper I, A Workshop Cipher - doi:10.5281/zenodo.21761191
- Paper II, Thirty Names per Sign - doi:10.5281/zenodo.21761982
- Dataset, Blind-card coding data of the chimeric-plant test (64 cards, one blind coder, sources of every image; CC BY 4.0) - doi:10.5281/zenodo.23205348
Read README.md first; every script is self-contained with a fixed seed. Questions,
reproduction reports, or a new candidate list to run through the pipeline - write to
a7777777vg1@gmail.com.
Third-party sources remain under their own licences and are linked, not redistributed. Our own corpora and scripts are shared for non-commercial research use.