Botanical Organization in the Voynich Manuscript Herbal Section: Cross-Section Anchor Density Analysis and the Three-Tier Folio Model

The herbal section of the Voynich Manuscript (f1r–f66v) has been variously characterized as a pharmaceutical index, a patient-routing directory, and a botanical reference in the medieval herbarium tradition. The HCIP research program (Hartlieb Cgm 7958 Iatromedical Patient Typology) proposed that the herbal section is organized by patient constitution class rather than by botanical or taxonomic principle, following the 28-mansion constitution typology documented in Johannes Hartlieb's Cgm 7958 (BSB Munich, 1456) [Honeycutt Ai Labs P-007, DOI 10.5281/zenodo.20392112; Honeycutt Ai Labs P-011, DOI 10.5281/zenodo.20391969]. We report the results of WP4, a cross-section anchor density extraction across 118 herbal folios (f1r–f66v) using the Zandbergen-Landini IVTFF v2b transcription as source. Five cross-section operators (otal, otedy, ytar, ary, qokaiin) [operator grammar per Honeycutt Ai Labs P-005, DOI 10.5281/zenodo.20391953] were extracted by exact-match token count across all herbal folios. Results: 73 of 118 folios (62%) carry zero hits across all five operators; the mean total anchor density is 0.79 per folio; no herbal folio matches or exceeds the f58r WP4-extraction total of 10. Four prespecified threshold tests (not statistically independent: they are three transforms of the same right-skewed distribution plus a cluster-pattern check) all fail for the operationalized constitution-organization hypothesis (H_const-as-five-anchor-density). The narrow operationalized form of the constitution-organized herbal hypothesis is REJECTED; the broader claim that the herbal is not constitution-organized in any encoding scheme remains beyond the scope of this five-operator test. Under the null/alternative frame, H_bot is not rejected by these data. The one herbal folio with elevated total anchor density in the WP4 five-operator extraction is f58r (total=10), which is well above the section mean and isolated at the late-herbal transition region. The empirical singularity within the WP4 dataset is established; the "preparation bridge" interpretation is auto-correlated with the baseline (the f58r=10 baseline was itself derived from f58r under the same WP4 extraction protocol in prior HCIP work) and is therefore presented as a model-internal interpretation, not externally corroborated. We note that the f58r=10 WP4-extraction value is a different operationalization than the registered f58r anchor count reported in Honeycutt Ai Labs P-017 v_real_1 (DOI 10.5281/zenodo.20649753), which uses the lattice assembly's high-signal anchor block (otaly×2, otal×4, okaiin×1; 7 entries) rather than the WP4 five-operator script; the two extractions overlap on otal=4 but otherwise count different token families, and the baseline calibration section of this paper documents the reconciliation. Interior herbal folios f48v (otedy=4), f43v (otedy=3), and f34r (otedy=3) carry elevated otedy token counts whose string form matches the f73v.23 <!09:30>otedy token (the labeled nymph at the 9:30 clock position on the Sagittarius zodiac ring) and the f84r label #5 token (the balneological output hub). The herbal-folio occurrences are token-matched only; they do not carry independent corridor attestation. The interpretation of these as intentional cross-references is therefore a candidate-tier reading, not a confirmed routing architecture. A descriptive three-tier partition of the WP4 distribution is proposed as a description of the observed five-operator counts (not a structural model independently predicted): (1) main botanical body (62% zero-density folios, conventional herbarium); (2) interior nodes with elevated otedy counts (f48v, f43v, f34r); (3) boundary cluster (f57v–f58r–f58v). Under the court physician workflow model, the rejection of H_const-as-operationalized is consistent with f58r operating as a translation point between herbal and balneological registers; this is expected under that model, not uniquely required by the WP4 data. Alternative models (compiled-volume, alphabetical, geographic, signature-doctrine, scribe-batch, quire-binding-order) are not tested here and are not excluded by the WP4 result.