The bifolium as a unit of layout in the Voynich herbal: root position and text block agree within a bifolium (preregistered automatic-mask test)

A preregistered blind test (doi:10.5281/zenodo.23219862 / 23228549) rejected the idea that roots in the Voynich herbal (Beinecke MS 408) were traced through the parchment. A post-hoc look showed that root positions of the front and back of a leaf agree better than chance without mirroring. Exploration on the blind root boxes: agreement holds for all four pages of a bifolium (honest null - bifolium labels permuted among leaves - p 0.0005; same in absolute pixels) and not for facing pages, neighbouring pages, other bifolia of the same quire or different quires. A preregistered test with an automatic ochre mask (#206; parameters frozen before the run) confirms it: roots z +3.72, p 0.0005; but the control - the text block in the upper half - agrees by bifolium almost as much (z +3.35, p 0.0015). Formulation by the preregistered rule: the whole folded sheet was laid out in one go (place of the plant and place of the text). The best-matching pairs show the same place but different root shapes - no copying of roots between leaves. Includes preregistration, code, masks, reports, inputs and figures. One coder for the exploratory part; the comparison with the codicological literature is not finished. Nothing here reads the text. Licence CC BY 4.0 (code also MIT). Page: https://voynich.site/roots-and-bifolia