Voynich | edkSG Research Volumes · EDKSG-VOY-V022 · Edition 1.0.0 · 5 September 2026, Singapore
Research status: executed fixed-coefficient support-concentration audit of the f105r split-policy strata inherited from Volume 13. This is a retrospective influence analysis over already-exposed evidence. No model refitting, source correction, population inference or decipherment is claimed.
No single family carries the sign.
The f105r split-policy page score is approximately +2.649. Its largest positive family, okeedy / qokeedy, contributes +1.420. Removing that entire family without refitting leaves the page at approximately +1.229.
The second-largest positive family, okedy / qokedy, contributes another +0.992. Removing both strongest positive families still leaves a narrow positive remainder of approximately +0.237.
Only after removing the next strongest positive family as well does the page cross below zero, to approximately -0.263.
The positive sign is therefore distributed across multiple supportive families, but most of its margin is concentrated in the two strongest ones.
1. Positive and negative families coexist
The page contains eight positive informative families and two negative informative families under the frozen coefficient, with two additional outcome-varying strata contributing zero difference.
Positive contributions sum to approximately +4.933; negative contributions sum to approximately -2.284. The published page score is the net of these opposing components.
2. The strongest supportive families
| Family | Contribution | Page score if removed |
|---|---|---|
| okeedy / qokeedy | +1.420 | +1.229 |
| okedy / qokedy | +0.992 | +1.657 |
| okeeody / qokeeody | +0.500 | +2.149 |
The leave-one-family-out result stays positive for every single family, positive or negative. That is a bounded robustness statement about the sign under one-family deletion, not evidence of independent replication.
3. Two-family concentration
The two strongest supportive families together contribute approximately +2.412. Removing both reduces the page to approximately +0.237.
So the sign does not hinge on either family individually, yet the available margin becomes small once both are removed. This is a useful middle state between “one family explains everything” and “support is evenly distributed.”
4. The third removal crosses zero
Removing the three highest-contributing positive families gives approximately -0.263. That operation is deliberately retrospective and influence-oriented. We are not proposing that these families should actually be deleted from the corpus.
Its purpose is to measure concentration: the page requires more than one family to sustain its positive sign, but not a large number of independent families.
5. Why this matters for source auditing
A source-lineage audit should prioritize the strongest supportive and adverse families, because uncertainty in them can move the page more than uncertainty in low-contribution families.
Volumes 20 and 21 already fixed a balanced audit list. Volume 22 adds a ranking principle: verify the two largest positive families and the two largest negative families before treating the page-level total as source-stable.
6. Influence is not correctness
A high-contribution family is not more likely to be historically correct. It is simply more consequential for this statistical claim.
Likewise, a negative family is not an error merely because it weakens the model. Source verification must be blind to whether a relation helps or hurts the preferred coefficient whenever possible.
7. The bounded conclusion
f105r’s positive split-policy score is not a one-family artefact. No single family deletion reverses it. But the two strongest supportive families carry most of the available margin, and a third high-ranked deletion reverses the page sign.
This makes the next source job narrower: audit the high-leverage families on both sides before expanding the model further.
Sources and reproducibility record
[1] Vol. 21 — The Aggregate-Cancellation Check.
[2] Vol. 13 — The Frozen-Coefficient Check. Its f105r split strata and frozen coefficient are used unchanged.
[3] EDKSG-VOY-V022-SUPPORT-CONCENTRATION-01. The accompanying package contains the parent strata, family contributions and leave-one-family-out ledger.
Collection: Voynich Research Library.
EDKSG-VOY-V022 · Edition 1.0.0 · 5 September 2026. Retrospective support-concentration audit. No decipherment is claimed.