An Operator Grammar for Beinecke MS 408: Full-Manuscript Validation, Challenge Testing, and Contemporary Manuscript Comparison

We define an operator grammar for Yale, Beinecke MS 408 based on compound reclassification of European Voynich Alphabet (EVA) transcription units into 31 operator classes, governed by 77 unattested transition pairs and system-wide positional constraints. The current E-run master dataset contains 109,514 intra-token transitions across 226 transcribed folio sides. An older report gave 114,507 transitions, but its parser and inclusion rules are not recoverable from the publication packet, so that value is treated as superseded rather than as a second validation frame. In the current E-run frame the only unattested-pair exception is a Zandbergen-Landini (ZL) inline-comment artifact at f107r, not a manuscript violation. The unattested-pair set was induced and checked on the same 226-folio corpus; held-out cross-folio validation is the next methods step. Twenty-one of 28 attested operators show greater than 80 percent positional bias in this corpus. Random sequences pass the combined constraints at 0.000 percent (0 of 10,000 trials); a Latin medical control text (Bartolomeo da Montagnana, Consilia Medica, Padua 1476) fails the eight specified positional comparisons. Operator frequency profiles predict Currier's Language A/B partition at 92.9 percent accuracy in a 30-feature cross-section model, distinct from an 88.3 percent result in a separate 7-feature boundary model. The Currier A/B labels and Davis hand labels are nearly co-linear in the ZL metadata (Table 6), so this accuracy is consistent with the 1976 partition but confounded with paleographic hand structure; it is not an independent confirmation of Currier 1976. Boundary detection recovers transition points correlated with Lisa Fagin Davis hand labels [1]; this supports structural co-variation between operator profiles and paleography, not a resolution of scribal identity. Challenge testing retained the structural findings within the tested controls while withdrawing the lexicon/triplet interpretation and downgrading specific Tironian semantic assignments to motivated hypotheses. A comparison with three contemporary Paduan pharmaceutical manuscripts (1355 to mid-fifteenth century) identifies 12 mark types across a century; distributional and manuscript evidence motivates operator-to-function assignments for 10 operators, but six discriminating computational tests did not distinguish these assignments from random at p less than 0.05, with one marginal label-content result at p equal to 0.0891. The operator-grammar reading is a structural finding, not a decoding. It is substrate-consonant with the recent statistical-semantic analysis of Layfield and Davis (2026) [2], the slot-grammar inducer of Zattera (2022) [3], and the entropy comparisons of Lindemann and Bowern (2020) [4]. Each reports structure inconsistent with its own stated comparator or null by a procedure distinct from the present approach; their test statistics, classes, and partitions are not interchangeable.