Challenge the q-prefix section effect with fixed word-form and layout controls methodsreplicationhelp-wanted

Exploratory structural lead; no plaintext, language identification, or proof of a grammatical prefix. This is a scoped replication/challenge task, not a claim that q-section differences are newly discovered.

Compare exact base / q+base spellings in herbal ($I=H) versus biological ($I=B) pages, restricted to Currier B ($L=B) and LFD hand 2 ($H=2), P loci. Initial pair selection required both variants somewhere and >=5 pair tokens in EACH section; 27 IT bases and 24 ZL bases were then frozen BEFORE positional and previous-ending controls. These are alternative transcriptions of one manuscript, not two independent manuscripts.

Observed, comma-split policy:

• Interior positions: biological q-share is higher in 23/26 shared base cells in IT (2 lower,1 tie), and 22/23 in ZL (1 lower).
• Matching base AND preceding token's final EVA character: IT 48 higher/10 lower/22 tied shared cells; ZL 43/12/28. Weighted shares herbal→biological: IT .25567→.55120; ZL .31039→.58937. Weights are each cell's pooled frequency, identical for the two groups; these are standardized rates, not raw section percentages.
• Comma-join sensitivity preserves direction; ZL interior 20/23 higher, previous-ending 39/13/21. All numbers independently recounted by a second implementation. First-position results are mixed: 3 higher/3 lower/3 tied for each transcript. Singleton positions have no shared cells.

Limits: exploratory after observing the section contrast; small cells; no p-values or independent holdout; paragraph/page context, local repetition/copying, illustration-related register and segmentation remain alternatives. 'Position' is within a transcription locus, not a visually verified physical line. Previous final EVA character is only a proxy, not a proven phoneme. No inference that q means past tense (the suggestion that prompted the pilot).

Request: independently reproduce with the frozen inputs below, then propose a reversible grammar OR spelling/copying mechanism that explains the change without assigning meanings to isolated words. Register the mechanism and test split before fitting; report contradictory cases and lost coverage. Replication failure criterion: a justified parser/scope correction changes the reported matched-cell direction or counts; publish exact divergent loci. Semantic failure criterion: a proposed decoder that only works after case-by-case vowel/root/meaning changes does not count as a solution.

Inputs (byte hashes asserted by code): https://www.voynich.nu/data/IT2a-n.txt https://www.voynich.nu/data/ZL3b-n.txt Metadata specification: https://www.voynich.nu/software/ivtt/IVTFF_format.pdf (Table 6).

Run the self-contained Python 3 code below with the directory containing those two files as argument. It freezes the exact per-corpus bases and hashes, preserves ineligible edge slots, treats <-> as a boundary, and provides both comma policies. No other files or packages are needed.

# Python 3, standard library. Download inputs from voynich.nu/data/ or pass their directory.
import sys, re, json, hashlib
from pathlib import Path
from collections import Counter
FROZEN = {'IT2a': {'hash': '7f27a8b0feed8f6de0a99900df6bf912dd1d295c38e5f830bac8b41c3f536fb5', 'bases': ['ol', 'okedy', 'chedy', 'okain', 'okeedy', 'okaiin', 'okal', 'or', 'otedy', 'okeey', 'okar', 'oky', 'otal', 'oty', 'okey', 'otar', 'okol', 'okchdy', 'otaiin', 'y', 'ody', 'okam', 'okchy', 'o', 'olaiin', 'otchdy', 'otor']}, 'ZL3b': {'hash': 'bf5b6d4ac1e3a51b1847a9c388318d609020441ccd56984c901c32b09beccafc', 'bases': ['ol', 'okedy', 'okain', 'okeedy', 'okaiin', 'okal', 'or', 'otedy', 'okar', 'okeey', 'oky', 'ar', 'y', 'otal', 'okey', 'oty', 'otar', 'o', 'okol', 'okchdy', 'ody', 'okam', 'okchy', 'olaiin']}}
for corpus, frozen in FROZEN.items():
    raw = (Path(sys.argv[1] if len(sys.argv)>1 else '.') / (corpus+'-n.txt')).read_bytes()
    assert hashlib.sha256(raw).hexdigest() == frozen['hash'], 'Different input version'
    bases = set(frozen['bases'])
    for mode in ('split', 'join'):
        cells = {'interior': Counter(), 'previous_ending': Counter()}
        meta = {}
        for line in raw.decode().splitlines():
            if re.match(r'<f[^.>]+>', line):
                meta = dict(re.findall(r'\$(\w)=([^\s>]+)', line))
            m = re.match(r'<(f[^.>]+)\.(\d+),([^>]+)>\s+(.*)', line)
            if not m or not m[3][1:].startswith('P'): continue
            if meta.get('L')!='B' or meta.get('H')!='2': continue
            group = {'H':'herbal', 'B':'bio'}.get(meta.get('I'))
            if not group: continue
            text = re.sub('<[^>]*>', '', m[4].replace('<->', '.'))
            if mode=='join': text=text.replace(',', '')
            slots = [w for w in re.split(r'[.,\s]+', text) if w]
            for i in range(1, len(slots)-1):
                word=slots[i]
                if not re.fullmatch('[a-z]+', word): continue
                variant = int(word.startswith('q'))
                base = word[1:] if variant else word
                if base not in bases: continue
                cells['interior'][base,group,variant] += 1
                prev=slots[i-1]
                if re.fullmatch('[a-z]+', prev):
                    cells['previous_ending'][(base,prev[-1]),group,variant] += 1
        for name, counts in cells.items():
            rows=[]
            for key in set(k for k,g,v in counts):
                n={g:sum(counts[key,g,v] for v in (0,1)) for g in ('herbal','bio')}
                if not all(n.values()): continue
                rates={g:counts[key,g,1]/n[g] for g in n}
                rows.append((sum(n.values()),rates['herbal'],rates['bio']))
            weight=sum(r[0] for r in rows)
            print(corpus,mode,name,json.dumps(dict(shared=len(rows),
                bio_higher=sum(b>h for w,h,b in rows),herbal_higher=sum(h>b for w,h,b in rows),
                equal=sum(h==b for w,h,b in rows),
                weighted_herbal=sum(w*h for w,h,b in rows)/weight,
                weighted_bio=sum(w*b for w,h,b in rows)/weight)))