New Gorhambury·Baconian Ciphers

One-Day Report · Biliteral Cipher · September 5, 2026

Today's claim

Riverbank’s own 1916 photographs of the two letterforms were turned into a measuring instrument and pointed at a single page of the First Folio — the Prologue to Troilus and Cressida. It sorted every one of the eight measurable w on that page into exactly the groups Riverbank marked — having never seen it. And their own alphabet turns out to show no difference at all for four of the commonest letters in English.

findings may be superseded by later reports

The short version

In 1899 an American cryptanalyst named Elizabeth Wells Gallup said she had found a hidden message inside the First Folio of Shakespeare — not in the words, but in the shapes of the letters. Two slightly different italic typefaces, alternating, spelling out a binary code devised by Sir Francis Bacon.

In 1916 her employers at Riverbank Laboratories published a manual showing, letter by letter, what they held the two forms to look like. Photographs of individual letters, each on a ruled grid, with a written description beside it. That book has sat on the internet, free, for years.

Today's result in one sentence. Those hundred-year-old photographs were turned into a measuring instrument and pointed at the Prologue to Troilus and Cressida — and it sorted every one of the eight measurable w correctly, into exactly the groups Riverbank marked, having never seen that page before.
The two sources, in full, free. Everything on this page comes from two documents anyone can open right now. The Keys for Deciphering the Greatest Work of Sir Francis Bacon (Riverbank Laboratories, 1916) — the manual that shows, letter by letter, what Riverbank held the two forms to look like; the exemplar plates used here are its pp. 38–72, and the table of borrowed letters is its p. 79. And the Bodleian’s photograph of First Folio, Arch. G c.7 — specifically the leaf carrying the Prologue to Troilus and Cressida, which is the one page tested throughout. Every glyph measured here was cut from it.
Bacon's bi-formed alphabet, each letter cut twice in two founts, with the five-place binary code
drag sideways on a phone · tap or click for full size
Start here if none of this is familiar. A bi-formed alphabet is an alphabet cut twice, in two slightly different founts. Five letters of ordinary type spell one hidden letter, by which fount each was set in. This plate — Bacon’s own demonstration, redrawn by Spedding in 1857 and reprinted by Riverbank in 1916 — makes the two founts obviously different so the principle can be seen. The founts Riverbank claimed to find in the Folio are nothing like as different, which is the whole problem.
The eight lower-case italic w on the Prologue, sorted into the two groups Gallup marked
drag sideways on a phone · tap or click for full size
The entire claim, with no mathematics in it. These are every measurable lower-case italic w on the test page — the Prologue to Troilus and Cressida — at full size, straight off the Bodleian’s photograph, sorted into the two groups Gallup marked in 1899. The question is whether the five on the left share something the three on the right do not. Everything else on this page exists only to answer a second question: could a sorting this good happen by luck?
The two stencils cut from Riverbank's 1916 exemplar plates
drag sideways on a phone · tap or click for full size
Riverbank's two forms of the letter w, laid on top of one another. Grey is ink they share and it decides nothing. Blue is ink only the a-form has; red only the b-form. Throw the grey away and six separate places to look are left.
All sixty pairs of Riverbank exemplars, XOR'd
drag sideways on a phone · tap or click for full size
The same operation on all sixty pairs in the manual, sorted most-different first. Lowercase w is the only one whose stencils, applied to the page, separate what Riverbank marked. Four letters — e, o, a, r — leave nothing to measure at all.
Separation for six letters under three different registration methods
drag sideways on a phone · tap or click for full size
A signal that is real should not depend on how the two shapes were laid on top of one another. Three fitting methods were tried — sliding only, then a nearest-point fit, then a nearest-point fit that also fits the letter’s lean. Lowercase w moves by 0.08 sd across all three. Capital P moves by 2.42 and f by 1.43, which is how we know those two were artefacts of the fitting and not properties of the page.
All sixty stencil pairs, coloured by the Folio passage each exemplar was cut from
drag sideways on a phone · tap or click for full size
Riverbank’s exemplar letters are not Bacon’s specimen alphabet. They were cut out of the First Folio itself and mounted on graph paper — and the book prints a table saying, for every borrowed letter, which passage, which line and which word. Six of the sixty were cut out of the Prologue: the page this project tests. Those six can never be scored on it. Lowercase w is not among them.
Separation against sample size for nineteen letters, with each letter's null envelope
drag sideways on a phone · tap or click for full size
The specificity control. One stencil — the one cut from Riverbank’s plates for w — applied to every letter on the page. If it were reading ink weight, or position in the line, or anything else the whole page shares, the letters with the most glyphs would show it first. e, with ninety-nine glyphs, comes out at 0.01. Eighteen letters land inside their own null envelopes. The nineteenth is w.
Agreement with Gallup's marking against sample size, for eight letters, from unsupervised clustering
drag sideways on a phone · tap or click for full size
Riverbank’s plates set aside entirely. A clustering program was shown only the glyphs, told nothing about ciphers or which letter was which, and asked to sort each letter into two groups. Its answer for w is Gallup’s 1899 answer, exactly — the same five together, the same three together. For e, with ninety-nine glyphs, it finds nothing that corresponds to her marking at all.

Questions

What is the biliteral cipher?

Sir Francis Bacon described it in 1623. Every letter of the alphabet becomes a group of five symbols, each symbol either an a or a b — binary code, three centuries before binary code.

The two symbols need not be letters at all. They can be two founts of type so similar that a reader notices nothing. Print an innocent passage in an alternating mixture of the two, and the shapes of the letters carry a second message underneath the first.

What was done today, in plain terms?

Riverbank's manual contains photographs of each letter in both of its claimed forms. Take one such pair, lay the two photographs on top of each other, and ask where they disagree.

Most of the letter is the same in both — that grey area tells you nothing, so throw it away. What is left is two stencils: the ink only the first form has, and the ink only the second has. By construction they cannot overlap.

Then take a real letter off the Folio page, lay it on the stencils, and ask which one it fills. That is the whole test. No judgment, no eye, no interpretation.

Why does it matter that the stencils come from their book?

Because every previous version of this test derived the two shapes from the page itself, and a fair critic could answer: you found a split and then named it.

This time the two shapes come from a book printed in 1916 and never touch the Folio until the moment of scoring. So do the scale and the alignment. Nothing about the instrument knows anything about our page.

There was one thing left to fix: the plates and the page are photographed at different sizes, so a conversion factor is needed. It was set from six letters — a e o d n s — chosen precisely because Riverbank's own plates show no difference at all between their two forms of those letters. A ruler cut from letters that carry no signal cannot smuggle any in.

And the result?

The five w Riverbank marks a all scored positive. The three it marks b all scored negative. No overlap, no letter near the boundary.

There are 56 ways those eight letters could be divided five-and-three. Riverbank's is the best one. Not among the best — the best.

Five other letters on the same page had enough clean examples to test the same way. The strongest of them managed about a quarter of w's separation.

You keep finding things against the claim. Is this one for it?

Yes. It is the strongest result this project has produced, and it should be said plainly rather than buried under qualifications.

What it establishes is narrow but real: on this page, the letter w comes in two shapes, those shapes match the ones a 1916 manual described in words, and a mechanical procedure can tell them apart without any human judgment. None of that was known yesterday.

What did their own alphabet give away?

Something they cannot have intended. Sixty pairs of exemplars came out of the manual. For four of the commonest letters in English — e, o, a and r — there is no difference to find. Lay their two photographs on top of each other and almost nothing is left over. Their own plates supply nothing to tell the forms apart.

That is their evidence agreeing, from the opposite direction, with the fifty-six measurements this project ran last week and published as null. We looked at the page and found nothing for most letters. Their book shows there was nothing to find.

So is the cipher real?

Not established, and today does not establish it. Two founts is not a message. A cipher needs the whole alphabet, and most of the alphabet still shows nothing — on the page and, it now turns out, in their own plates.

What has changed is the instrument. Until today, testing Riverbank's claim meant reconstructing what they might have meant. Now their stated forms can be measured directly, at any scale, on any page, by anyone with the same free files.

What happens next?

Eight letters is a small foundation. The same procedure now runs unchanged on more letters and more pages, and there are about 900,000 letters in the Folio, every one of them free to fetch.

Three things are queued. Whether the split survives on the rest of the cancelled leaf, which nobody claims is enciphered. Whether other letters separate once there are hundreds of examples rather than eight. And whether the whole alphabet, scored this way, produces anything but noise.

Any of those can kill it. That is the point of running them.

Who is doing this and why?

Walter Wilkinson, Worland, Wyoming. The sister strand is the Word Cipher work published alongside this.

The reason it has merit is that the question has stayed arguable for 127 years purely because nobody measured it — and until about six years ago, nobody easily could. The photographs went online free. The statistical machinery got built for another purpose entirely. Both halves now exist, and no one had put them together.

The wider project lives at New Gorhambury. Worth knowing before you read it: those pages argue a conclusion. These reports deliberately do not.