Enki Sumerian
We are measuring the world’s oldest writing as pure geometry, wedge by wedge, to see whether it is organised like a coordinate system.
By Brad Harris, whose bio and field notes explain why base-60.
Enki runs on a base-60 coordinate system. The oldest writing we have counted in 60s.
Enki gives every node a permanent base-60 address, and its collapse equation folds many signals into one governed node. The address is only a handle; the meaning lives in the library it points to.
Cuneiform was pressed into wet clay about 4,000 years ago, and its makers counted in 60s, the system still behind our minutes and seconds. So we asked a narrow question: if you ignore what the signs are said to mean and only measure their shapes, are they built and used like addresses?
Every sign becomes a list of wedges, and every wedge four numbers
Distances are in bits: one bit is the width of a standard wedge head, so every sign sits on the same ruler.
U+12427, redrawn from its 20 measured wedges. Every sign on this page is drawn this way, from our numbers, not from a font.
Five checks stand between the raw sources and anything on this page
Sources are pinned and hashed
Every downloaded file has a SHA-256 fingerprint in a manifest. The files these results depend on were re-checked against it on 18 October 2026 and match exactly.
Only signs actually on the clay are kept
Editors mark signs they restored or supplied where the clay is broken. Those are dropped, and damaged lines are left out of the line tests, so nothing guessed enters the counts.
Every wedge is confirmed by eye
A detector proposes wedges; a person then checks each sign on a grid, moving, adding or removing wedges until it is right. Signs that do not fit the wedge model are held out, not forced.
Every pattern is tested against chance
Each result is compared with the same data shuffled: the same numbers in random order, the same signs on other tablets, or random sets of numbers. A pattern counts only if it beats the shuffle.
Everything can be rerun
The scripts, their outputs and the measurements are public. When the scripts were rebuilt from scratch they reproduced every number exactly, and anyone can rerun them.
What we have seen so far
Four observations, each with what it could mean and what it does not show.
Signs are built like counters
Seen252 of 541 signs are one small unit repeated a set number of times. We call each family a ladder. The clearest is a diamond of 4 wedges, found 2 to 9 times.
Diamond ladder: 2 to 9 diamonds of 4 wedges each
Down-wedge ladder: 1 to 9
Hook ladder: 1 to 9
On the tablets, hooks stop at 5 before units
SeenAcross 673,286 runs of counting signs, a hook sign directly followed by down wedges has 1–5 hooks 58,383 times and 6–9 hooks only 11 times. The 6–9 hook signs exist and are used 1,422 times elsewhere. Down wedges after hooks run all the way to 9.
Uses of each hook sign, by what comes right after it. Dashed line: between 5 and 6. The 10-hook sign behaves differently and is left out.
Lines repeat like stored records
Seen78.9% of complete lines appear symbol for symbol on another tablet. One 15-symbol layout appears 4,395 times on 584 tablets: five fields, each a number followed by two signs, where the first sign repeats and the second changes.
N = a number. Tinted letters repeat from field to field, like a network shared by several machines.
Looking alike does not mean being used alike
OpenFor 64,261 pairs of common signs we compared how different they look with how alike their neighbours are in the text. Overall there is no link at all (rank correlation -0.001). Signs on the same ladder, though, share neighbours 2.6 times as much as unrelated signs that differ by the same amount.
Where Enki’s coordinate system comes in
Each Enki idea becomes a question the tablets can answer.
| In Enki | Question for the tablets | What the data shows so far | Status |
|---|---|---|---|
| Every node has a permanent base-60 address | Are signs built from fields with fixed ranges? | Units run 1–9; hooks stop at 5 when units follow (58,383 uses against 11) | Seen |
| Collapse: many signals become one node | Do different-looking signs behave as one thing? | Signs on the same ladder share their neighbours 2.6× more than unrelated signs, but most ladders are numbers, so this may only mean numbers act like numbers | Open |
| The address is a handle; meaning lives in the library | Does looking alike predict being used alike? | No: rank correlation -0.001 across 64,261 pairs | Seen |
| Governed records, reused exactly | Do whole lines repeat like stored records? | 78.9% of complete lines recur symbol for symbol on other tablets | Seen |
| Missing values can be inferred from the rules | Can unseen signs be predicted? | 64 gaps predicted; 4 found, all weak | Not seen |
| Codes guard against misreading | Are signs kept far apart? | No: 7 identical pairs and 84 signs one wedge from another | Not seen |
Seen: the data shows it, against chance. Open: shown, with a caveat that could explain it away. Not seen: the data does not show it.
What we do not know
The drawing might be the font’s, not the writing’s
Every sign was measured from a modern font. A pattern could come from how its designer drew the signs. Measuring the same signs on 3D scans of real tablets is the test.
Sign order comes from modern editions
No translations are used, but which sign stands where was decided by editors. If they tidied what they saw, patterns like the 5-hook limit could partly be theirs.
What the place above the hooks is worth
Hooks stop at 5, which points to 60, but the sums test scores every value from 20 to 120 almost the same. It is not settled.
The 10-hook sign
It is the most-used hook sign (60,794 uses here) and stands before down wedges 4,129 times. It does not fit the 1–5 ladder.
Numbers behave like numbers
Signs on the same ladder are used alike, but most ladders are counting signs, which all sit in number positions. The library idea has not yet been tested outside numbers.
Only part of the library is checked
541 signs cover 94.1% of the text in this period. The font has about 680 more, almost never used in these tablets.
Curved marks
Three signs contain a curved cup outline, the only non-wedge marks in the set. They may be a font convention and are left out until checked on clay.
A few counts may be off by one
Signs were confirmed by eye. A handful sit one wedge off a clean pattern (marked * on this page) and are queued for a re-check.
What counts against the idea
- Symbol use is very uneven: 970 symbols, and the top 100 make up 83.2% of the text. Designed codes usually spread use more evenly.
- 7 pairs of signs are identical by our measure, and 84 signs are one wedge from another sign. A misread wedge would change the value.
- Slanted wedges cluster at 45° and 135° (442 against about 124 by chance), not at multiples of 30°, which hold 211 against about 249. The circle is split in eighths, not sixths.
- Predicting missing signs mostly failed: 4 of 64 gaps were filled, all weakly.
Where every input comes from, and how we checked it
| Source | What we use it for | Licence | How we checked it |
|---|---|---|---|
| ePSD2/CDLI Ur III Corpus (Oracc JSON build of 7 Dec 2022) Oracc (University of Pennsylvania) with the Cuneiform Digital Library Initiative | Which sign stands where in each line, on 79,660 tablets. No readings or translations. | CC0 (stated in the download’s metadata); Oracc site pages are CC BY-SA 3.0 | SHA-256 re-verified against our download manifest on 18 Oct 2026; site reachable |
| Oracc Global Sign List (OSL) Oracc | Links each sign ID in the corpus to its Unicode code point. Nothing else is taken from it. | No licence stated in the repository; used as a lookup, not republished | Pinned to commit 6231cc4 (8 Oct 2026); repository reachable |
| CuneiformClassic font (Uni12000Cuneiform 2.001) Steve Tinney and Michael Everson | The standard drawing of each sign, measured to find its wedges. The font is never shown or shared; every sign on this page is redrawn from our own measurements. | © 2011–2012 Steve Tinney and Michael Everson, all rights reserved | SHA-256 re-verified on 18 Oct 2026; copyright read from the font file |
| Unicode Cuneiform code charts The Unicode Consortium | Code points such as U+12423 are used as neutral labels for signs. | Unicode terms of use | Charts reachable (U+12000 and U+12400 blocks) |
| HeiCuBeDa Hilprecht 3D tablet scans Heidelberg University (heiDATA), published 6 Jun 2019 | Downloaded for the next step: checking the font-based measurements against real clay. Not used in the results on this page. | CC BY-SA 4.0 | DOI resolves; licence read from the dataset record |
Also downloaded but not used in these results: the CDLI bulk text dump, SumTablets, the MTAAC corpora and ETCSL. They will be listed here if a later result depends on them.
What we test next
- Libraries outside numbers. Group signs only by their neighbours in the text, then check whether each group shares a shape field.
- Field ranges. In repeated record layouts, check whether each field has its own limit.
- The place above the hooks. Look for counts that run 58, 59 and then roll over.
- Clay against font. Measure a sample of signs on 3D scans of real tablets.