Voynich | edkSG Research Volumes · EDKSG-VOY-V026 · Edition 1.0.0 · 7 September 2026, Singapore
Research status: executed prospective page-choice alternate-source reconstruction of f77r using Takeshi Takahashi’s H transcription after selecting the page before inspecting its H-based score. The frozen model, event definition and conditional scoring rule remain unchanged. This is not a historically blind experiment, independent manuscript replication, image adjudication or decipherment.
The second full-page result goes the other way.
The retained ZL-derived f77r page contributes approximately +2.545 under the frozen common coefficient. The full Takahashi H reconstruction contributes approximately +4.046.
That is an increase of approximately +1.501, or about 159.0% of the retained-source positive gain.
The increase does not come from a page-wide uniform strengthening. It comes entirely from one exact current-form family: ol/qol. Its conditional contribution rises from approximately +2.472 to +3.973. Every other informative family’s frozen contribution is unchanged to numerical precision.
The source-lineage effect is therefore directional but not monotonic across pages: Takahashi weakens f105r and strengthens f77r. That makes transcription choice a structural source of variance, not a one-way correction toward or away from the model.
1. Why f77r was selected before the H score
After Volume 25, the next clean test was to apply the same whole-page alternate-source procedure to another fixed-model target without first looking at its alternate score. f77r was chosen because it had already been one of the three Volume 13 evaluation pages and because its dominant family had been source-lineage-audited locally in Volume 19.
The page choice is therefore prospective relative to this H-score calculation, but not historically blind in a stronger sense. The manuscript page, baseline score and prior local audits were already known. We preserve that distinction.
2. The Takahashi page is reconstructed as one whole text unit
The historical display identifies forty paragraph-text lines for Takeshi Takahashi on f77r. The page also contains separate labels and nymph names; these are not part of the paragraph-text experiment.
Three paragraph blocks are recovered from line-final paragraph markers. The same inherited parser and boundary extractor used in Volume 25 then operate on the forty text records.
The H reconstruction yields 115 admitted transitions, compared with 107 in the retained split-policy f77r event index. Yet both representations contain exactly 34 informative observations. More admitted rows again do not imply more conditional information.
3. The page score increases materially
| Representation | Admitted transitions | Informative strata | Informative observations | Frozen common gain |
|---|---|---|---|---|
| Retained ZL-derived f77r | 107 | 5 | 34 | +2.545 |
| Takahashi H reconstruction | 115 | 4 | 34 | +4.046 |
The alternate representation contains one fewer informative exact-form stratum but a larger page score. The missing stratum in the retained representation had essentially zero model-versus-zero contribution, so its disappearance does not explain the increase.
4. One family accounts for the entire increase
The retained ol/qol family has counts n=9, r=6, m=6, k=6. Its frozen common contribution is approximately +2.472.
Under the Takahashi reconstruction, the same exact-form family has counts n=13, r=7, m=7, k=7. Its contribution rises to approximately +3.973.
The difference between those two family contributions is +1.500794. The entire f77r page-score difference is also +1.500794.
This exact accounting matters. A larger page total does not mean that every family became more supportive. One family expanded into a more perfectly aligned configuration while the others retained their previous conditional contributions.
5. The dominant family is more strongly separated in H
In the H representation, all seven predecessor-y observations in the ol/qol family are q-prefixed outcomes, while all six non-y predecessors are unprefixed outcomes.
The retained representation also shows perfect alignment inside its smaller nine-observation version of that family: six y→q and three non-y→unprefixed cases.
The alternate transcription therefore does not create a different direction in the family. It increases the number of represented cases exhibiting the same separation.
This is not thirteen independent confirmations. Several occur on the same page, some on the same manuscript paragraph, and all refer to one historical object.
6. The other informative families do not drive the change
The oteedy/qoteedy family remains adverse at approximately −0.801. The olcheedy/qolcheedy family remains positive at approximately +0.500. The okaiin/qokaiin family remains mildly positive at approximately +0.374.
A near-zero okedy/qokedy stratum present in the retained page does not remain informative in the H reconstruction, but its disappearance changes the page score only at floating-point noise.
The score increase is therefore concentrated and fully auditable.
7. The hand-specific frozen model also rises
With the retained hand-2 analytical routing label held fixed, the frozen partial-sharing model scores approximately +4.549 on the Takahashi reconstruction.
That exceeds the common model’s +4.046 on this page. As with f105r in Volume 25, one alternate-source page is not enough to reopen model selection. The ordering is recorded, not promoted.
8. Boundary-policy equality needs careful interpretation
Split, join and exclude-touching policies return the same H score on this reconstruction.
This does not establish that the manuscript has no uncertain word boundaries. It means that the H page representation used here does not expose the same doubtful-comma pattern that made those policies diverge in the retained ZL-derived analysis.
Once again, identical sensitivity totals are related views of one representation, not independent replications.
9. f77r and f105r now give opposite source-lineage movements
This is the most important comparative result of Volumes 25 and 26.
On f105r, replacing the retained page representation with Takahashi reduced the frozen common score from approximately +2.649 to +1.273.
On f77r, the same style of alternate-source reconstruction increases the score from approximately +2.545 to +4.046.
A transcription lineage is therefore not acting like a simple correction factor that always weakens or always strengthens the predecessor-y clue.
Source representation changes the available exact-form comparisons in page-specific ways.
10. What this does—and does not—strengthen
The result strengthens one bounded statement: the positive predecessor-y/q-prefix association on f77r is not dependent on using only the retained ZL-derived transcription. A complete Takahashi reconstruction produces a larger positive score.
It does not make the overall model twice as credible. The two transcriptions are not independent manuscripts, the page was already known from earlier development, and the increase is dominated by one exact-form family.
It also does not resolve whether the H or retained segmentation better reflects the physical manuscript. That remains a source/image question.
11. The next experiment should move beyond favourable-page selection
We now have two complete alternate-source page results with opposite score movements. A natural next test is the third Volume 13 target, f34r, chosen before inspecting its H result.
That page contributed much less positive evidence in the retained evaluation. Including it prevents the alternate-source series from becoming a sequence of only the strongest pages.
After all three are reconstructed, the correct aggregate analysis must again decompose page-level and family-level changes rather than reporting only a single total.
12. Bounded conclusion
The full Takahashi reconstruction of f77r remains positive and is materially stronger than the retained ZL-derived result, but the increase is entirely concentrated in the ol/qol family. Together with the f105r decrease, this shows that transcription-lineage sensitivity is real, substantial and page-specific.
No semantic value has been assigned to q, y, ol, qol or any other represented form.
Sources and reproducibility record
[1] Volume 25 — The Full-Page Alternate-Source Check.
[2] Volume 13 — The Frozen-Coefficient Check, supplying the frozen models and retained f77r baseline.
[3] Voynich Information Browser — f77r, exposing the Takeshi Takahashi H paragraph-text rows and the Currier metadata for the page.
[4] Landini–Stolfi interlinear archive lineage and IVTFF conventions as recorded in LSI_ivtff_0d.txt, the complete archive used in Volume 25.
[5] EDKSG-VOY-V026-PROSPECTIVE-ALTERNATE-PAGE-01. The package contains the H-row ledger, deterministic wrapper, frozen parser/model code, complete policy outputs, dominant-family accounting and nine regression checks.
Collection: Voynich Research Library · Editorial framework: Wintour House.
EDKSG-VOY-V026 · Edition 1.0.0 · 7 September 2026. Prospective page-choice alternate-source sensitivity audit. No independent manuscript replication, image adjudication, source correction or decipherment is claimed.