Evidence · Voynich Manuscript

Read through Greshko's segmentation, a working two-letter solver finds no reading

With the segmentation fixed by Greshko's prefix and suffix inventories, a solver that recovers two-letters-per-word keys finds no reading of the manuscript.

Suggestive0

What it means

Key questionWhat kind of text is it?

Most likely: a notation or memory system

Most likelyunchanged

Part of itIs it a cipher of a known language?

Partly answered

Most likelyunchanged

Answer this evidence bears onEach word is a fragment of a verbose cipher

Counts againstnone for, 2 against

Disputedunchanged

Next steps 5

T39, second round of the section-level grid: add the two mechanisms that produce what T38 cannot reach. (a) Re-spacing: write the cipher as a continuous glyph stream and place spaces by a rule (after word-final glyphs such as -y, -n, -r, -l, before q-, ch-, d-), so word edges depend on neighbouring glyphs (T15's re-spaced stream gave edge MI 0.22). (b) Local copying: with probability p the next word is a one-glyph variant of a word among the last w (self-citation), for the near-twins. Cross with the T38 table-stay knob and the line layer, confirm best settings on fresh seeds.

~8 h

On: Is it a cipher of a known language?

—not judged

Naibbe-type key over glyph strings learned from the manuscript rather than Greshko's inventory (which reads 78 % of tokens), with whole-word signs for common words, validated on planted keys first. It would also have to keep page locality, which letter-level homophones erase (M-T36).

~16 h

On: Is it a cipher of a known language?

—not judged

Codebook attack: group one-glyph variant families that share contexts, then align the manuscript's co-occurrence structure with Latin texts of the same genres (unsupervised lexicon induction: embedding alignment, graph matching). Validate on a planted Latin codebook with variants laid into the same pages first.

~20 h

On: Is it a cipher of a known language?

—not judged

Section-level key search: run the M-T37 solver within single sections (training and held-out pages alternating by bifolio), so a key need not hold across sections whose order and scribes differ.

~6 h

On: Is it a cipher of a known language?

—not judged

Ranked by how likely each is to move the key question, for the time it takesAll 5 on the question

Sources 2

The Naibbe cipher: a substitution cipher that encrypts Latin and Italian as Voynich manuscript-like ciphertext

G 2025

SRC

greshko/naibbe-cipher: Python implementations of the Naibbe cipher and Voynichesque, decryption script and figure data

Greshko 2025

SRC

Cite it

Mysteries Department. (2026). Read through Greshko's segmentation, a working two-letter solver finds no reading [Evidence EVD0DC7RD2B8AP5]. https://mysteriesdept.com/r/EVD0DC7RD2B8AP5

@misc{mysteriesdept_evd0dc7rd2b8ap5,
  title     = {Read through Greshko's segmentation, a working two-letter solver finds no reading},
  author    = {{Mysteries Department}},
  year      = {2026},
  note      = {Evidence, catalog key EVD0DC7RD2B8AP5},
  url       = {https://mysteriesdept.com/r/EVD0DC7RD2B8AP5}
}