Writer’s guide
Begin with one group, then learn the proposed components and writing rules.
Start the guideBeinecke MS 408 · Working notes
Compare manuscript scans, Latin-letter transcription, and one working interpretation across thirty selected pages.
This is not a solved-manuscript claim. No language or cipher has been identified. Confirmed Voynich words and plaintext clauses remain zero.
01 / Start
Learn the proposed system, compare complete pages, or look up one copied group.
Begin with one group, then learn the proposed components and writing rules.
Start the guideSee the scan, raw text, and proposed reading for each of 30 pages.
Browse the pagesSearch all 2,051 copied spellings and open any occurrence in context.
Look up a group02 / Updates
Source and presentation changes, newest first. Publication times are UTC.
GDT672 gives each of the 28 lines on f1r a cautious experimental reading. All 214 copied groups remain visible: 129 directly reuse a V48 card. The other 85 use local cards made for this page—54 composed from existing parts, 23 learned as whole forms, and eight tied to one occurrence. Seventeen quantity links are stated explicitly, and no group is left without a card. The copied page and all other page readers are unchanged. The prose still uses 85 broad carrier terms, so substantial uncertainty remains visible. This is a working proposal, not plaintext or a translation.
GDT671 assigns provisional cards to 24,159 of 32,339 positions in the transcription and leaves 8,180 open. In this model, 1,242 lines with several groups have no blanks; 318 also pass its stricter check. Its files contain 1,430 glossary forms, 2,115 technical dictionary rows, and 56 proposed roles. V48 does not change the copied text or the other page readers, and it identifies no language, sound system, plaintext, plant, or ingredient.
The separate material-pattern model advanced to V38. GDT661 assigned provisional cards at 872 previously open positions in the Latin-letter transcription. Forty-six cards applied only to an exact complete group; standalone r and d used context-specific readings instead of universal meanings. The model filled 18,451 of 32,339 positions and left 13,888 open. It had 233 lines with several groups and no model blanks, 99 under its stricter internal filter. These were not confirmed words or translations, and the complete reader for the 30 selected pages remained unchanged.
The writer’s guide kept the experiment separate from the complete 30-page view. A short summary stayed visible; six recurring examples and the context-specific r/d explanation opened on demand. The 556-group glossary and 678-row technical table were available as complete downloads.
V27 filled 14,617 of 32,339 transcription positions with working proposals and left 17,722 open. The model was later revised and extended through GDT661; GDT651 replaced the three sheckh… readings.
Two independent readers and a separate reconciliation checked five paired entries from two historical herbal manuscripts. The pairs supply concrete fields for a later comparison. The earlier synthetic decoder path was closed because its constraints could not be satisfied. No Voynich target page was opened, and confirmed Voynich words and plaintext clauses remained zero.
GDT599 supplied working proposals for 313 paragraphs on six pages, while the other 480 paragraphs retained GDT594 and all 744 local cards retained GDT589.
Every raw group remains visible by default. Each complete raw paragraph now sits directly beside its complete current interpretation, so the two can be compared without scrolling between separate sections. Technical source wording and alternatives remain available below each pair.
GDT594 supplied all 793 running paragraphs, while GDT589 supplied all 744 local cards. The later composite reader retains GDT594 on 480 paragraphs.
Every raw group became visible by default, with technical evidence kept optional.
Four pages brought the selected-page view to 30 pages and 5,866 groups. The state-card layer also separated 687 complete working operations from 19 narrower roles. No word or plaintext sentence was confirmed.
GDT513 completed the separately labelled local layer for the earlier 26 pages: 183 groups in 135 short records, 501 component defaults, and nine section markers. No word or sentence was confirmed.
Hover, tap, or focus a group to connect its copied spelling with proposed pieces, meanings, and context.
This 26-page snapshot made immediate context the final safety check while retaining its proposed readings.
This 22-page edition contained 4,581 copied groups and 627 proposed sections.
Each selected folio received its own scan, raw transcription, and proposed-structure view.
A trial reading was removed when later source revisions no longer supported its components and grouping.
Passes 1008–1010 expanded the edition, joined nearby groups into sections, and reviewed three proposed grades.
The guide introduced a short method followed by a searchable component reference.
Selecting a raw group opened its proposed parts and meanings beside the line.
Four pages and 657 groups were added without establishing a reusable sign meaning or plaintext.
Repeated proposed component meanings were separated from interpretations that depend on nearby drawings.
This pass introduced a 159-part component key, 70 proposed combinations, and readings for 2,511 groups.
The first public edition presented fourteen page readings and their limits.