5/5
Known structures
recovered first
91–93%
Bars before a 76-word
against 33% expected
70%
Of T's vocabulary in the Staff
against 39–54% elsewhere
6
Tablets carry formula F1
no other skeleton > 4
Contents AbstractWhat is claimedValidationFour results In the seriesRevisionsStated limitsCite
// abstract

Rongorongo has been read many times and measured rarely. This paper measures it. It asks what the structure of the Santiago Staff (text I) and the Honolulu tablet (text T) can establish before any sign is given a meaning. Every structural claim is tested against a null model that keeps the signs and destroys their order, and the pipeline is first required to recover five structures already established in the literature. It recovers all five. The structure that follows is that of a register without numerals, with an open identifier slot coupled to its head. It does not decide between an administrative register and the ta'u, the traditional list of names and titles; a list of names with titles is itself a register of persons.

Read on Zenodo Download PDF Code and outputs

// what is and is not claimed

Structure first, meaning never assigned

Rongorongo, Rapa Nui
Texts auditedSantiago Staff (I) and Honolulu tablet (T); the verso of the Small Santiago tablet (Gv) as the comparison case
CatalogueBarthel's (1958) numeric sign catalogue
SourcesThree transcriptions, none by the author: the rongopy tablet transcriptions (de Souza), the Santiago Staff transcription distributed with de Souza (2025), and a derived compilation of the kohaumotu.org transcriptions
Null modelKeeps the signs of each text and destroys their order
RobustnessEvery test run on two transcriptions where two exist, with damaged signs both removed and kept
Does not attemptDecipherment. No sound or meaning is assigned to any sign, and the paper does not argue that rongorongo is not writing.
// validation

The pipeline had to find what was already known

Before any new result was reported, the same code and settings had to recover five structures established by earlier researchers.

Negative control. The central test was run on 2,508 words of connected Rapa Nui text, taking each of the eight most frequent particles as the delimiter. None is spaced more regularly than chance. The regular spacing found on the Staff and T is a property of those inscriptions, not of the test.

// four results

A template, a head, a pair and a formula

// in the excel in clay series

A register without numerals

The earlier papers read Linear A, the Indus script and Proto-Elamite as structured records. For Rapa Nui, only the rows that were tested here are filled in.

Texts I and T, slot by slot
Entry headX.76, opening each unit; 163 distinct heads in the Staff — tested
Entry templateRegular unit length; second slot coupled to the head — tested
Section dividerCarved bars, grouping 1–18 units — tested
Standard formulaF1, outside I and T, with a variable field — tested
NumeralsNot found — tested negatively
Closing operatorNot identified — not tested, not claimed

The comparison runs in one direction only. It shows that the Staff has the architecture of a record. It does not show that the Staff records what a Minoan or Elamite archive recorded. Because the Staff does not count, its audit had to be built from structure and null models alone — a method that can now be applied back to the earlier papers in the series.

// a note on revision

What the tests took back

The paper lists the claims its own tests withdrew, among them:

Language models helped write the analysis scripts and were used adversarially to look for errors. No output was accepted because a model asserted it, only because it agreed with the data — the verification architecture argued for in No Argument for the World and developed as Reference Signal Engineering.

// stated limits

What this audit does not establish

Every result is a result about transcriptions

No sign was read from the wood. All transcriptions descend from one catalogue (Barthel 1958). The Staff results agree on two transcriptions, but these are not independent readings of the object; Horley's (2021) tracings are the obvious next source.

T carries a claim larger than its size

T has 114 legible words, 147 of 261 transcribed words are flagged as damaged, and only one transcription was used for the tests. T has the Staff's template on every test that T is large enough to take.

The genre is not decided

The structure fits a register of persons as well as an administrative register. An exploratory calibration is reported with its limitations, and a fixed protocol for deciding the genre is specified in advance for version 2.

// access and citation

Open access, permanent identifier

Related: Excel in Clay I · Excel in Clay II · Excel in Clay III · Concepts glossary · All seven audits

Diener, Ł. (2026). Excel in Clay IV — The Rapa Nui Register: A Null-Tested Structural Audit of Rongorongo Texts I and T (version 1.0). Zenodo. https://doi.org/10.5281/zenodo.23030356

Preprint, not peer reviewed by a journal. Licensed CC BY 4.0. The deposit includes the analysis scripts, a data-retrieval script that verifies every input against its commit or SHA-256 checksum, and the full output of every run reported in the paper.

Read on Zenodo Excel in Clay I Excel in Clay III All seven audits Zenodo community