The Voynich Discriminant Zone
Among 55+ corpora from 35+ language families, only Tagalog (Noli Me Tangere, canonical full-corpus) enters the 95% confidence interval for Voynich VMML. But the match collapses when a second metric is introduced.
The VMML proximity between Voynich and Tagalog (Noli) is the closest approach by any natural language tested. The Boundary Concentration gap — 0.161 units — represents a decisive structural mismatch. No corpus in our 55+ collection simultaneously enters the Voynich zone on VMML, BC, and CBMI. The Discriminant Zone is empirically empty.
The Alphabetic Ceiling
Every alphabetic natural language tested falls at or below VMML = 5.748, derived from Nietzsche's Also sprach Zarathustra — the highest-scoring text in the initial 35-corpus scan. Voynich exceeds it.
The ceiling is empirically derived, not theoretically imposed. German was the anchor candidate because its compound morphology places it near the top of the natural-language VMML distribution. The ceiling has been replicated across multiple language samples. The sole exception is Tagalog (Noli Me Tangere, full corpus) — which is the subject of Finding 3.
Tagalog VMML is text-specific, not a language property
The supra-ceiling VMML observed in Tagalog (Noli Me Tangere) does not generalise. The same author's second major novel yields a VMML below the ceiling — ruling out the interpretation that Tagalog as a language exceeds the alphabetic ceiling.
Both texts are by José Rizal, written within four years of each other, in the same language and literary register. A cross-text VMML delta of 0.336 units — larger than the ceiling margin itself — demonstrates that Tagalog VMML is highly sensitive to text-level factors: vocabulary, topic, sentence structure. This instability disqualifies the Noli result as evidence of a language-level property.
Structural mechanism divergence
Voynich and Tagalog achieve similar VMML values through fundamentally different internal mechanisms — confirmed by character permutation testing. This is the strongest evidence that VMML proximity alone is not sufficient to claim structural similarity.
The permutation test randomly shuffles characters within each corpus (n=50 trials, corpus-size matched at n=19,968 tokens) and measures the resulting VMML. A larger decay signals that more of the original VMML depended on specific character ordering — that is, on real internal structure. Voynich decays more than twice as much as Tagalog. The CI95 ranges do not overlap. Same VMML level, opposite structural mechanisms.
Focus-morphology density gradient hypothesis
A working hypothesis emerging from the Austronesian analysis: VMML elevation above the alphabetic ceiling may correlate with focus-morphology density rather than simple phylogenetic membership in the Philippine branch.
The evidence: Tagalog (complex focus system, Noli) and Ilocano (provisionally, n=12,807) exceed the ceiling. Cebuano (NLLB corpus, likely code-switching bias) does not. Indonesian and Malay — Austronesian but without Philippine-type focus morphology — fall well below the ceiling (4.838 and 5.293 respectively). The hypothesis predicts that formal Cebuano, and Kapampangan and Hiligaynon, should exceed the ceiling if focus-morphology density is the causal variable. These tests are pending. This finding is explicitly flagged as a working hypothesis, not a confirmed result.
The discriminant zone remains empty
After 63 systematic corpus tests across 35+ language families, no natural language simultaneously occupies the Voynich zone on all three metrics. The zone is not empty because we haven't looked — it is empty after exhaustive testing.
VMML alone narrows the field to one candidate (Tagalog, Noli, full corpus). BC eliminates that candidate. CBMI provides further separation. The three-metric combination defines a zone that no tested natural language occupies. This is the operational definition of our central claim: Voynich belongs to a structural class not represented in the current comparative baseline. The claim is falsifiable — testing new corpora could fill the zone. It has not been filled in 63 attempts.