RΛIN’s Research Atlas / Manuscript imaging / Observatory trail 3 of 3
Voynich manuscript images: what the infrared view reveals
The same part of a page can look strikingly different under different capture conditions. In this folio 8r comparison, most visible writing becomes less distinct in the infrared-labelled capture, while one dark patch remains prominent. Its origin and the disputed reading remain unresolved.
Research direction: RΛIN / ATO. Exploratory analysis and public summary developed with AI assistance, 4 November 2026. This page summarizes STAR-INK-06 and STAR-INK-07; it is not independently peer reviewed. The images are existing 2014 captures, not new imaging undertaken by ATO.
One location. Four views.
Look at the end of the word, toward the right of each crop. These are the same upright pixel region, shown without reconstructing or drawing new strokes.
Image credit: The Lazarus Project and The Chester F. Carlson Center for Imaging Science at Rochester Institute of Technology. Manuscript: Beinecke Rare Book & Manuscript Library, Yale University, MS 408, f. 8r.
Each monochrome crop has its own 1st–99th percentile contrast stretch. Brightness is therefore not a calibrated comparison between channels. The supplied capture labels are retained; exact filter response curves were not independently established.
Does the pattern come from the other side?
The wider infrared image contains a plant-shaped pattern. The follow-up mirrored Yale’s ordinary-color f8v scan and aligned it with eight manually selected plant landmarks away from the word. Repeated leaf and stem arrangements support the broad pattern as the reverse illustration showing through the leaf.
The small patch is a separate question. Both an affine fit and a projective sensitivity fit place its predicted reverse-side location on apparently bare parchment between leaves. No obvious corresponding dark mark was identified in that visible scan. That does not establish recto-side ink: a weakly visible reverse feature, stain or other material could respond differently in infrared.
What the result supports
| Question | Current result |
|---|---|
| Broad plant pattern from the reverse illustration? | Supported by mirrored geometric correspondence. |
| Origin of the isolated dark patch? | Unresolved. No obvious visible reverse counterpart was identified. |
| Extra stroke, different ink or a particular reading? | Not established by these images. |
| Decipherment or new linguistic meaning? | Not established. |
How close is the alignment?
With each landmark left out in turn, prediction RMS error was 73.7 pixels for the affine model and 54.3 pixels for the projective model, in the full-resolution recto frame. The largest omitted-landmark errors were 133.0 and 78.4 pixels. Those errors are large relative to a small stroke: the alignment supports a broad pattern comparison, not stroke-level certification.
The two full fits predict reverse-side positions about 7.2 verso pixels apart. Agreement does not eliminate systematic errors shared by both models. Full images were inspected before the landmark protocol was recorded, so this is exploratory and unblinded.
The next discriminating observation
A comparable reverse-side infrared capture, or a more accurate local alignment based on features outside the disputed strokes, could help separate the patch’s possible sources. A blinded reading comparison can test visible alternatives, but it cannot establish material origin on its own.
Why this belongs beside the 4D question
In the yin–yang investigation, a precise observation rule lets a failed candidate be rejected. In the pairwise projection lab, different hidden relations share the same summaries. Here, changing the capture reveals how one manuscript image can combine several physical contributions. The connection is a method for asking better questions; manuscript imaging supplies no evidence for a 4D yin–yang or a wormhole.
Sources and exact crop
Lisa Fagin Davis’s release announcement gives the imaging history and required credits. Source material: public imaging archive, f8r raw captures, and Yale f8v color scan, image 1006091.
Crop bounds are [3060, 4600, 3430, 4750), measured in the upright 6132 × 8176 frame. The raw arrays were rotated 90° clockwise without interpolation. Embedded historical TIFF paths name “08v” despite the current folder, filenames and visible layout identifying f8r; this discrepancy is retained in the audit. The original reports preserve the raw files, hashes, alignment protocol and replay code.
Explore the connected research map → · Restart the three-stop trail
You must be logged in to post a comment.