New Gorhambury·Baconian Ciphers

One-Day Report · Biliteral Cipher · September 7, 2026

Today's claim

Riverbank’s other two 1916 booklets were read plate by plate. In their own words, the alphabets were redefined until the groups decoded, and doubtful letters were settled “by the assistance of the context”. The one numerical rule they ever printed for telling the two forms apart — a lean of ten degrees against fifteen — was measured on the very page it was written for, and fails; every other rule in the booklet fails with it. The lower-case w still stands alone, taller than every chart, and a defect in its second-page test is corrected here in the open.

findings may be superseded by later reports

The short version

In 1899 Elizabeth Wells Gallup said the First Folio carried a hidden message in the shapes of its italic letters: two founts so alike that no reader notices, alternating to spell a binary code Bacon had described in 1623. In 1916 her employer, Riverbank Laboratories, published three booklets explaining how she did it. One of them, The Keys, has been this project’s instrument since the second of September. The other two — The Greatest Work of Sir Francis Bacon and Hints to the Decipherer — were read this weekend, from the plate images, page by page.

They turn out to say two things this project has been trying to establish by measurement. First, how the forms were found: by assigning letters, checking whether the group of five decoded, and when it did not, redefining the forms. Second, how the forms are to be told apart: a list of rules, one of them a number. Every rule that can be turned into a measurement was then measured on the Prologue to Troilus and Cressida — the page Riverbank chose — against the marks Riverbank printed.

Today’s result in one sentence. Riverbank said the b-form leans about five degrees more than the a-form; on the page they marked, across fourteen letters, the difference is a tenth of a degree, and the wrong way — and none of their other rules for size, tails, dots, counters, serifs or baselines follows their marks either. Only the lower-case w separates, as before, and by a feature their rules never name.
The sources, in full, free. The Greatest Work of Sir Francis Bacon (Riverbank, 1916, 28 pp.) and Hints to the Decipherer (Riverbank, 1916, 24 pp.), both from Library of Congress copies, readable page by page at New Gorhambury; The Keys for Deciphering (Riverbank, 1916) on archive.org; and the Bodleian’s photograph of First Folio, Arch. G c.7, the leaf carrying the Prologue. Every glyph measured here was cut from that leaf; every mark was taken from a table Riverbank printed in 1916.
For fourteen letters, the difference in lean between b-marked and a-marked glyphs, against Riverbank's predicted five degrees
drag sideways on a phone · tap or click for full size
Riverbank’s own rule, measured. Hints to the Decipherer (1916, p. 12) says the a-form letters lean about ten degrees from the perpendicular and the b-forms about fifteen. Each dot is one letter on the Prologue: the average lean of the glyphs Gallup marked b, minus the average of those she marked a. The gold line is the five degrees the rule requires. Nothing reaches it; nothing comes close.
Every italic e on the Prologue, sorted by lean, with the eight roman e's set aside
drag sideways on a phone · tap or click for full size
What “every italic e on the page” means once the page has been cleaned by hand. Ninety-nine glyphs had been labelled e by the machine; fifty-four are whole italic e’s, eight are roman e’s from a different fount, one is broken, and thirty-six are other letters or fragments. The fifty-four lean between 16° and 25° in a single hump, the a-marked and b-marked alike. The eight roman ones, upright with a level bar, are the likeliest source of a “second hump” in any angle histogram that has not set them aside.
Lean distributions for the letters of ELIZABETH, one dot per glyph, coloured by Gallup's mark
drag sideways on a phone · tap or click for full size
The letters of ELIZABETH, one dot per glyph. Two founts differing in lean would show as two humps whether or not the colours followed them. No letter has two humps; height and the distance each letter sits above its baseline — two more of Riverbank’s rules — do not follow the marks either. z occurs once on the page.
Effect sizes for Riverbank's rules on ink weight, tails, serif exits and last strokes, letter by letter
drag sideways on a phone · tap or click for full size
The rest of the rules, first half: ink weight, the tail of e and c, the exit stroke of i, n, u, the last stroke of n. Each row is one letter and one measurement; the dot is the b-marked minus the a-marked, in units of the letter’s own spread. A usable difference would sit at the right-hand edge. Forty-five measurements in all; two reach p < 0.05, which is what chance produces in forty-five.
Effect sizes for Riverbank's rules on the i-dot, counters, top serifs and the t crossbar
drag sideways on a phone · tap or click for full size
Second half: the position of the i-dot, the counters of a, d, o, e, the top serifs of l, d, h, and the cross-stroke of t. The t crossbar is the same length to three decimals whichever way the letter was marked.
The stress axis and contrast of e, o and a against the marks, with w's height on the same scale
drag sideways on a phone · tap or click for full size
For scale. The top rows are one more measurement — the angle at which the punchcutter’s pen was held, read from where each bowl is thinnest — and it too ignores the marks. The last row is the lower-case w’s height, the audit’s one positive, drawn on the same scale: it stands at two, off the right edge of every chart above. That is what a usable discriminator looks like, and it is the only one this page has yielded.
The eleven w's of the Digges poem sorted by height, coloured by a blind two-way split
drag sideways on a phone · tap or click for full size
The w on a second page, corrected. On 5 September this test was reported as a pass with a modest correlation; it had been run on glyphs the binarisation had broken into pieces. Taken whole, the eleven w’s of the Digges poem fall into seven small and four large with no overlap in height — visibly the two sorts seen on the Prologue — while, against the exact wording registered in advance, width edges out height as the strongest correlate. A technical fail and a stronger result, both stated.

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 over these two days, in plain terms?

Two of Riverbank’s 1916 booklets had not been read. They were fetched from New Gorhambury, where they are kept as page images, and read from the plates rather than from any machine transcription — a rule this project adopted after a text layer put words into Riverbank’s mouth last week.

One booklet, Hints to the Decipherer, contains the only list Riverbank ever published of how to tell the two forms apart. Each rule on that list that can be turned into a number — lean, height, position above the baseline, ink weight, the direction of a tail or a serif, the position of the dot on an i, the shape of a counter, the length of a crossbar — was measured on every verified glyph of the Prologue and compared with the mark Gallup gave that glyph. The marks come from a table the other booklet prints: the whole Prologue in groups of five with its a/b code, which agrees with the marking this project has used since the second of September in all 1,070 places.

What do Riverbank’s own books say about how the forms were found?

More than a critic would dare to put in their mouth. The Greatest Work (p. 8): “Frequently a letter would be assigned during the examination to the a or to the b form only to find that such assignment resulted in a combination which was meaningless, when the group of five to which it belonged was compared with the key. Further examination and comparison were then of course necessitated, and a redefinition of characteristics of the respective forms followed.”

And Hints (p. 11), on Gallup’s early practice of leaving doubtful letters “to be accomplished later, by the assistance of the context”: “This plan, though legitimate, is, of course, unscientific, and too much open to adverse criticism to be satisfactory as anything more than a temporary expedient. It must be altogether eliminated from any permanent system of decipherment.”

That is the circularity this project has been measuring, stated in print by the people who did the work. The replacement they proposed for it is the quadrant rule.

What is the quadrant rule, and did it work?

“As a rule the a-form letters show an angle of about ten degrees between their chief upright lines and the perpendicular, while the b-forms show a slope of about fifteen degrees.” The decipherer was to equip himself with “a small quadrant or projector, a fine 4-inch rule, or some cards cut to right angles.” It is the one place in all three booklets where a number is offered.

It does not work on the page it was written for. Measured on the chief upright of every stem letter and on the long axis of every round one, the b-marked glyphs lean a tenth of a degree less than the a-marked, averaged over fourteen letters, and a third of a degree more by the median. No letter comes within reach of five degrees; for o and i the difference runs the wrong way. The scatter within a letter is two or three degrees, so a five-degree gap would have been unmissable. The rule Riverbank printed does not reproduce the marks Riverbank printed.

And the other rules?

The booklet lists eleven in all: the a-forms “larger, rounder, more upright, and more robust”; the tail of c and e curving in for one form and pointing out for the other; the serif strokes of i, m, n, u; long-tailed letters sloping more; circular letters judged by the slope of their longest diameter; the quadrant; letters standing above the baseline; the last stroke of m and n shortened; the position of the dot on i; a line through the counter of a, b, d, g, o, p; the top serifs of tall letters and the cross-strokes of f and t.

Ten of the eleven have now been measured on the Prologue and none follows the marks. Every e tail on the page curves in; the “pointing out” form does not occur. Every letter sits on its baseline to within a twelfth of a millimetre whichever way it was marked. Forty-five separate measurements produced two at the conventional threshold, which is what forty-five measurements produce by chance. The eleventh rule, on long-tailed letters, needs more glyphs than this page has.

Is anything still standing?

The lower-case w. On the Prologue its five a-marked glyphs are 45 pixels tall and its three b-marked 56, with nothing between: on the scale used for every chart above, an effect of two, where nothing else on the page passes one. It separates by height and by the shape of its counter — features that Riverbank’s written rules, as it happens, never mention.

And it holds on a second page, with a correction. The Digges poem’s eleven w’s, taken whole, fall into seven small and four large with no overlap in height. Two sorts of w in this italic is now the best-supported fact in the file. It is also exactly what a printing house with two cases of type would produce, and it says nothing about any other letter.

You found a mistake in your own earlier result. What happened?

The 5 September Digges test was run on a lower-resolution scan than the Prologue, and at that resolution the step that turns the photograph into black-and-white broke most w’s into two or three pieces. The pipeline then kept only the piece under the centre of each box — between a third and two thirds of the ink for nine of the eleven glyphs. The reported pass was computed on fragments.

It came to light while this report was being assembled, because the figure needed the glyphs drawn and the drawn glyphs were plainly incomplete. Re-run on whole glyphs, the split is cleaner than before but width edges out height as its strongest correlate, and the prediction registered in advance named height. So the honest record is: a technical fail on the registered wording, a stronger result in substance, and both runs kept. That is the format working as intended.

What would change your mind?

In either direction. A second letter that separates on the Prologue the way w does, by any rule at all, measured before anyone looks at the marks. Or the w split failing to appear on the Catalogue and Actors leaves once enough glyphs are found there. Or Riverbank’s per-passage Classifier plates, which have not yet been extracted, showing two forms for e where their descriptive plates show one.

What would not change it is another way of reading the marks. The marks are fixed; they were printed in 1916 and confirmed twice.

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. The photographs went online free; the statistical machinery got built for another purpose; and it turns out the claimants themselves left a list of what to measure. The wider project lives at New Gorhambury, and its Experiment One. Worth knowing before you read them: those pages argue a conclusion. These reports deliberately do not.