Voynich | edkSG Research Volumes · EDKSG-VOY-V024 · Edition 1.0.0 · 6 September 2026, Singapore
Research status: executed receiving-family propagation of the two reading changes isolated in Volume 23. The reassigned f105r.16 occurrence is retained in its mapped oteedy/qoteedy family rather than treated as lost. The coefficient remains frozen. This is bounded page accounting, not a complete alternate-transcription rerun, image verdict or decipherment.
The reassigned occurrence does not disappear.
In the retained ZL3b-derived representation, f105r.16 contains oiir → okeedy. In the Takahashi-attributed display used by Volume 23, the same passage is oiir → oteedy. Volume 23 correctly removed that occurrence from the selected okeedy/qokeedy family, but stopped before asking what happens in the family receiving it.
The retained f105r event index already contains four observations in the oteedy/qoteedy family. Their frozen-model contribution is approximately +0.219. Adding the reassigned non-y → unprefixed oteedy occurrence raises that receiving-family contribution to approximately +0.378, a change of +0.159.
When that receiving-family effect is propagated alongside the two Volume 23 reading changes, while every other f105r stratum is held fixed, the bounded page accounting is approximately +1.926.
The local four-family subtotal reverses in Volume 23, but once the reassigned occurrence is restored to its receiving family, the bounded f105r page accounting remains positive. The reading differences materially weaken the page; they do not, under this limited propagation, reverse it.
1. The accounting gap left deliberately open
Volume 23 selected four high-leverage families before inspecting the historical display. One mapped reading changed a current form from okeedy to oteedy. That occurrence therefore no longer belonged inside the selected four-family projection.
The previous article preserved the moved event in an out-of-family ledger and explicitly refused to pretend it had vanished. That was the correct stopping point because its receiving family had not yet been reconstructed.
A sign change in a selected subtotal can be real while still being incomplete accounting for the page. Family reassignment requires two entries: subtract the occurrence from its original family and add it to the family it joins under the alternative representation.
2. The receiving oteedy/qoteedy family already exists
The retained split-policy f105r index contains four events whose exact current-form family is oteedy/qoteedy.
| State | n | Predecessor-y | q-prefixed outcomes | Overlap | Frozen contribution |
|---|---|---|---|---|---|
| Retained family | 4 | 3 | 1 | 1 | +0.219 |
| After adding f105r.16 oiir → oteedy | 5 | 3 | 1 | 1 | +0.378 |
The new member is a non-y predecessor followed by an unprefixed current form. Under the fixed positive coefficient, that alignment is favourable to the represented association and increases the family’s conditional contribution.
This is not new manuscript evidence. It is the same occurrence represented in another documented transcription display and moved to the family implied by that display.
3. The source mapping is explicit
The current repository object records the relevant line as yteody.oteeeos.aiin.odal.oiir.okeedy.oral. The compared historical display records yteody oteeeos aiin odal oiir oteedy oral. The only change propagated for this occurrence is the current form okeedy → oteedy. The preceding oiir remains non-y and the current form remains unprefixed.
That source difference changes an exact-form identity without changing the two binary features by itself. Its statistical consequence arises because exact current-form identity determines which conditional stratum receives the observation.
4. The second Volume 23 change remains separate
At f105r.17, the compared display changes sheokedy → qokedy to sheoked → qokedy. That occurrence stays in the okedy/qokedy family, but its predecessor changes from y-ending to non-y.
Volume 23 already propagated that predictor change inside the focal family. Volume 24 does not alter it. The new work here is solely the receiving-family side of the f105r.16 reassignment.
Keeping the two mechanisms separate matters: one observation changes strata; another changes predictor class within a stratum.
5. Bounded page accounting
The original f105r split-policy frozen score is approximately +2.649.
Across the four high-leverage families, Volume 23 changed the subtotal from +0.127 to -0.755, a difference of -0.882.
The receiving-family propagation contributes back +0.159.
original f105r score +2.648786
four-family mapped change -0.882042
receiving-family change +0.158814
----------------------------------------
bounded mapped accounting +1.925559The page remains positive in this bounded accounting.
6. Why +1.926 is not a complete historical-source page score
The value is useful precisely because its scope is explicit. It propagates the two source differences already isolated in Volume 23 and restores the one reassigned occurrence to a known receiving family. It does not discover or propagate every difference between the complete ZL3b-derived page and a complete historical raw transcription.
The Takahashi-attributed comparator used by Volume 23 is a displayed space-normalized page, not the complete original annotation stream with all uncertainty and alignment syntax. Other families could gain, lose or change observations under a full source-compatible reconstruction.
Accordingly, +1.926 is a bounded mapped accounting, not a replacement for Volume 13’s published page score and not a corrected manuscript estimate.
7. Family reassignment must conserve observations
This experiment exposes a general research-control rule. When a reading change moves an observation from one analytical category to another, sensitivity accounting should preserve the observation unless the new representation makes it genuinely ineligible.
Subtracting it from the old category without testing the destination can exaggerate the impact of the reading change. Forcing it to remain in the old category can hide the impact. The correct operation is a traceable transfer.
That principle is particularly important in exact-form conditional models because one character can change the matching family while leaving broader binary features unchanged.
8. What this changes in our understanding
Volume 23 correctly showed that the high-leverage four-family balance is reading-sensitive. Volume 24 shows that part of that local reversal is an artefact of looking only at the departure side of a family reassignment.
The two documented reading differences still weaken the f105r result substantially in the bounded accounting: approximately +2.649 becomes +1.926. That is a reduction of about 0.723 log-likelihood units.
But the sign remains positive after conserving the reassigned occurrence.
The appropriate claim is therefore neither “the historical reading reverses f105r” nor “the reading differences do not matter.” They matter materially, and the bounded propagated result remains positive.
9. The next research boundary
The next step should not continue patching one selected cohort indefinitely. The useful next threshold is a full-page alternative-source reconstruction with an explicit cross-source boundary policy, or a sufficiently detailed image audit of the two influential disputed readings.
Until one of those is available, this branch has reached a natural local stopping point: original family, changed reading, receiving family, predictor change and bounded page consequence are all explicitly accounted for.
The wider predecessor-y programme remains subject to the source-lineage and representation qualifications accumulated through Volumes 15–24. Nothing here supplies a semantic value for q, y or any represented word.
Sources and reproducibility record
[1] Vol. 23 — The Balanced-Reading Check, EDKSG-VOY-V023. Its fifteen-case mapping and out-of-family record are retained unchanged.
[2] Vol. 13 — The Frozen-Coefficient Check, providing the original f105r split-policy event index and frozen coefficient.
[3] ZL3b-derived f105r diplomatic object, including the retained f105r.16 line and the existing oteedy/qoteedy observations.
[4] The Takahashi-attributed f105r display identified and bounded in Volume 23, used only for the two already documented mapped readings.
[5] EDKSG-VOY-V024-RECEIVING-FAMILY-01. The accompanying package binds Volume 23’s result ledger to the original split events, reconstructs the receiving-family counts and reproduces the bounded page accounting with seven regression checks.
Collection: Voynich Research Library · Editorial framework: Wintour House.
EDKSG-VOY-V024 · Edition 1.0.0 · 6 September 2026. Receiving-family propagation audit. No complete alternate-transcription page score, image adjudication, independent replication or decipherment is claimed.