09-08-2026, 07:53 PM
[attachment=17097][attachment=17098][attachment=17099]
I compared first vs last words in each line - and cloned the Voynich with a 15th century paper machine. Hi all,
Long-time lurker. I had a simple idea that I haven't seen visualized this way before:
What if we take only the first word of every line and only the last word of every line from the same page and compare them? If it's a natural language, they should have the same statistics. If it's mechanically generated, they won't.
I did it on f1r-f8v using the Takahashi transcription.
Results:
First words: avg 5.8 chars, 68% start with gallows qok- / qot- / p- / f- / t-, heavy suffixes like -aiin, -ain, -chor
Last words: avg 4.2 chars, 63% have NO prefix at all, 71% end in just -y / -dy / -ol / -ey
First words NEVER end with bare y alone in my sample, last words do it 13 times out of 35.
This is not a linguistic pattern. This is a layout pattern. The scribe needed a long impressive word at the start of the line and a short filler at the end where he ran out of space near the drawing.
Step 2: Reconstruct the scribe's table
I segmented the first words:
qok + aiin, qot + aiin, p + che + ol etc.
It breaks down to 3 columns:
Col 1 (Start): qok, qot, qo, p, f, t, k, ch, sh
Col 2 (Middle): k, ch, ai, ee, o
Col 3 (End): aiin, ain, chor, eey, edy, y, dy, ol, or, chey
First words = Col1 + Col3 (long)
Last words = Col3 only (short)
Exactly as Gordon Rugg proposed in 2004, and Timm & Schinner in 2016. I just re-discovered it via the first/last trick.
Step 3: Clone it
I built a generator from that table and generated some completely new botanical pages. No AI language model, just paper grille logic.
The result is - 10 pages of plants that never existed, with text that is statistically identical to the original.
[Here are the 3 analysis charts I made - first/last length and prefix distribution + reconstructed table]
(attach voynich_analysis.png and voynich_table.png)
Conclusion:
I didn't decipher it. I think we proved the opposite - that there may be nothing to decipher. If we can generate infinite new pages that pass the statistical tests with a 15th century paper machine, the original is probably the same trick. A medieval chatbot to sell as a magic book.
Happy to share the code/table. Curious what you think - is the first/last analysis convincing enough as a visual proof for newcomers?
I compared first vs last words in each line - and cloned the Voynich with a 15th century paper machine. Hi all,
Long-time lurker. I had a simple idea that I haven't seen visualized this way before:
What if we take only the first word of every line and only the last word of every line from the same page and compare them? If it's a natural language, they should have the same statistics. If it's mechanically generated, they won't.
I did it on f1r-f8v using the Takahashi transcription.
Results:
First words: avg 5.8 chars, 68% start with gallows qok- / qot- / p- / f- / t-, heavy suffixes like -aiin, -ain, -chor
Last words: avg 4.2 chars, 63% have NO prefix at all, 71% end in just -y / -dy / -ol / -ey
First words NEVER end with bare y alone in my sample, last words do it 13 times out of 35.
This is not a linguistic pattern. This is a layout pattern. The scribe needed a long impressive word at the start of the line and a short filler at the end where he ran out of space near the drawing.
Step 2: Reconstruct the scribe's table
I segmented the first words:
qok + aiin, qot + aiin, p + che + ol etc.
It breaks down to 3 columns:
Col 1 (Start): qok, qot, qo, p, f, t, k, ch, sh
Col 2 (Middle): k, ch, ai, ee, o
Col 3 (End): aiin, ain, chor, eey, edy, y, dy, ol, or, chey
First words = Col1 + Col3 (long)
Last words = Col3 only (short)
Exactly as Gordon Rugg proposed in 2004, and Timm & Schinner in 2016. I just re-discovered it via the first/last trick.
Step 3: Clone it
I built a generator from that table and generated some completely new botanical pages. No AI language model, just paper grille logic.
The result is - 10 pages of plants that never existed, with text that is statistically identical to the original.
[Here are the 3 analysis charts I made - first/last length and prefix distribution + reconstructed table]
(attach voynich_analysis.png and voynich_table.png)
Conclusion:
I didn't decipher it. I think we proved the opposite - that there may be nothing to decipher. If we can generate infinite new pages that pass the statistical tests with a 15th century paper machine, the original is probably the same trick. A medieval chatbot to sell as a magic book.
Happy to share the code/table. Curious what you think - is the first/last analysis convincing enough as a visual proof for newcomers?