To the eye, an alphabet is pure history. Scripts are borrowed, sounds drift, and every culture bends the letters to fit its own tongue. Nothing about it looks like mathematics.
Yet underneath, the names of the letters form a hidden structure. Follow it, and an exact number appears: the Golden Ratio squared. And it appears not once, but in writing traditions that never met.
THE DISCOVERY
We took 36 writing traditions, from Hebrew and Arabic to Thai, Korean, Runic and English, each with its own set of letter names. We applied one strict rule to all of them: follow how each letter’s name refers to other letters, and measure how that network grows. We used no tuning and did not go looking for famous numbers. Every tradition got the same rule.
The results were not random. They gave a small family of exact constants:
φ² = 2.618… the Golden Ratio squared
Hebrew • Phoenician • Ge’ez • Thai • Japanese-pronounced Latin
2 + √2 = 3.414…
Aramaic • Syriac • Burmese
1 + √2 = 2.414…
Arabic
And English? Its letter names land on 3.141593…, a number that agrees with π to five digits. It is not π itself, but it is close enough to make you look twice.
IT IS NOT LUCK
We built 4,482 random networks of the same size and shape. The Golden Ratio squared appeared in only 5.8% of them. The chance that three unrelated writing traditions all land on that exact number by accident is less than 1 in 100.
SEE IT FOR YOURSELF
Three Hebrew letters, Alef (א), Dalet (ד) and Lamed (ל), are tied together by their names. Each one points to the others. Written as a grid, the loop looks like this:
[1 0 1]
[1 1 1]
[1 1 1]
Three letters. One loop. Out comes φ² = 2.618…, the Golden Ratio squared.
Nobody tuned this. These are just letter names, the same ones used for three thousand years.
THE MYSTERY AND THE OPEN CALL
The data is exact, and anyone can check the math. But why this happens is still unknown. The blind tests have carried the work as far as one researcher can.
Break it. Try other historical spellings and pronunciations. Does the pattern survive?
Find the next one. Test an alphabet nobody has tested yet. Which number does it give?
Explain it. Why do human letter names keep choosing the same few exact numbers? We need a theorem.
WHERE IT COULD MATTER
Mathematics and symbolic dynamics. Which exact numbers can human naming systems produce, and why so few? A theorem is waiting to be written.
Combinatorics on words. Every alphabet becomes a sequence that generates itself. What kind of sequences are they?
Linguistics and the history of writing. A numerical fingerprint for each script: a new way to compare traditions, test transliterations and trace how alphabets descend from one another.
AI and language technology. Tokenizers that split text for AI models follow the same kind of substitution rule. Could this fingerprint predict how well a language compresses? Can a model learn the rule from the strings alone?
Information theory. The growth rate is a single number that measures how much hidden structure a naming system carries.
Cognitive science. Why would people naming letters, centuries and continents apart, build the same small loops?
Physics and biology. Do the same loops and constants appear in models of growth, or in how biological codes are named?
These are open questions, not claims. Each one is a paper waiting for the right person.
The framework has now survived five out-of-sample tests: it replicated under an inherited naming tradition, collapsed exactly where the filter predicted (Vietnamese), matched a timestamped prediction (Ogham), exposed a false positive in a lineage’s modern descendant (Glagolitic vs Cyrillic), and produced a target constant on command (√3). We do not claim a law. We claim a machine that has not failed a test yet, and we are inviting you to try to break it.
We are saying something stranger:
the names of the letters themselves follow rules no one has measured until now.
WHAT IS ALREADY ON THE TABLE
Open dataset. The full incidence matrices for all 36 writing traditions, as CSV files, with a verification script that recomputes every growth rate in seconds.
https://doi.org/10.5281/zenodo.23440579
Technical paper. Exact factorizations for every script, the small core loops behind each constant, an exhaustive baseline of 4,482 random networks, a significance test corrected for shared history, a complete classification of the small quadratic constants, five out-of-sample blind tests, open conjectures, and full references.
https://doi.org/10.5281/zenodo.23449728
I am looking for collaborators, co-authors and institutional partners
to take this to the next scale.
Contact: Bahaa Budargham • ORCID 0009-0005-2955-3703
The door is open.
