Tested K&A at scale, found something, need your help

[CorwinFr]

I need to understand what you mean exactly first.

Good question, and it made me recheck the claim properly. Short answer: the 53% is an artifact. I'm retracting it.

How it was computed

Take the first word of each herbal folio, strip the longest suffix from a fixed list (ol, or, ar, al, dy, am, etc.), keep roots ≥3 chars. That gives 108 unique roots. Then check if any of those roots appear as substrings inside the 286 pharma block openers. Result: 158/286 = 55%. Original null test against random VMS words gave p=0.002.

Why it's wrong

The null test drew random words from all positions in the VMS. But block-initial words have a very different character distribution: 81% of herbal roots and 87% of pharma openers start with a gallows (p, t, k, f). When I control for this and draw random block-initial words instead:

Code:

Real (herbal roots):          158/286 = 55% Null (random block-initials): 183/286 = 64% z = -2.03, p = 0.944

Herbal roots match fewer pharma openers than random block-initial words. The overlap comes from shared gallows+digraph patterns in block-initial position, not from a herbal-pharma link.

The claim is retracted. README, paper, and GitHub updated: You are not allowed to view links. Register or Login to view..

The 7 other structural properties and the fingerprint identifications are unaffected, they use separate methods with separate null tests.

I'll be stepping away from Voynich work for a while. Need to stop working on this at 2am. Thanks for pushing on this, it's exactly the kind of scrutiny