Home Research Theory AI & PCT Why AI Lies Audit Kit Consulting Cases Blog FAQ About
1,916
Deduplicated
inscriptions
584
Unique signs
11,110
Sign occurrences
98.31%
Sequence uniqueness
Mohenjo-daro, n = 179
Contents CorpusFive fieldsNumerals Commit operatorMetrology VerificationStated limitsCite
// corpus

A primary key is not a sentence

Audited material · Harappan seals, c. 2600–1900 BCE
Inscriptions1,916 verified, deduplicated
Unique signs584
Sign occurrences11,110 tokens
Mean length4.6 signs
Sub-corpusMohenjo-daro, n = 179 texts
Sequence uniqueness98.31% (permutation test p < 0.001)
Conditional entropy3.23 bits
SourcesFuls ICIT database (2020) · Mahadevan Concordance (1977) · Parpola Corpus (1994) · Kriger & Hunt Mohenjo-daro sub-corpus (2026)
SanitisationAnti-MNS protocol — database cleaned of ISC/IVC tag contamination before any count was taken

A corpus in which 98.31% of sequences occur exactly once is not distributed like a language. It is distributed like a set of identifiers. Natural language reuses its sequences; primary keys do not, and neither does this.

// five fields

Frozen boundaries, a dynamic interior

The near-duplicate result is the one that closes the linguistic reading. If these were sentences, single-sign edits would fall anywhere. They fall in one place, every time, with a probability of coincidence under one in a hundred.

// the numeral anomaly

Not counts. Tariff flags.

Stroke numerals concentrate in field four: 45.71% of all stroke numerals in the corpus appear there, making up 15.41% of total corpus volume, across 154 numeral tokens.

Within that field a single stroke-2 sign accounts for 66.91% of numeral entries — 103 of 154 tokens — tested against a uniform distribution at χ² = 198.3, p < 0.001.

That is the decisive anomaly. If these were item counts, Benford's Law predicts digit-1 dominance. The corpus does the opposite, and does it overwhelmingly. A number that clusters on one value across seven centuries is not counting goods; it is recording an institutionalised rate.

// terminal operators

One sign, 1,395 times, always at the end

Sign 311/342 occurs 1,395 times — between 10% and 12% of total vocabulary volume — and converges on the terminal position. The audit reads it as a transaction-closing operator: applied to a clay sealing, it confers the status transaction verified and closed, duty paid.

Its absence is the finding that makes the reading testable. Without the terminal key the document is not a valid transport record but an unapproved draft — a draft/approved distinction encoded positionally rather than lexically. A second sign behaves as the functional analogue of the Minoan ki-ro and Sumerian lá-NI deficit operators, which are independently attested in their own corpora.

// metrology

The jar and the brick were built to the same standard

Convergent measurement system
Reference weight13.63 g, with binary fractions 1/2, 1/4, 1/8
Verified fits3.41 g = 0.25 exactly · 6.8 g = 0.5
Stroke arithmeticShort shallow strokes encode a base-8 sub-system; long strokes enable multiples of 6 and 12
Storage jar target9.47 L
GeometryCube of edge 21.156 cm = 12 Indus Angula at 1.763 cm
Observed deviation0.6% – 3.9%, attributed to thermal shrinkage in firing
Architectural lock3:4 ratio with the 28.2 cm Harappan brick (16 Angula)

A container volume that resolves to a whole number of units of the same length standard used for construction is not a coincidence of pottery. The paper calls it a harmonic lock: logistics and architecture running on one metrology.

// formal verification

The schema was machine-checked before it was interpreted

The pipeline is public. This is the part of the audit that a disagreeing reader can rerun rather than argue with.

// stated limits

What the numbers do not settle

Entropy alone is not decisive

Conditional entropy of 3.23 bits indicates strong sequential structure but is, on its own, compatible with both a constrained non-linguistic code and a natural language. It is reported as supporting evidence, not as proof.

Which identifications are attested and which are hypotheses

The Minoan and Mesopotamian operator parallels are independently attested in their own corpora. The Harappan identifications remain the audit's hypotheses, rendered structurally plausible by the convergence rather than demonstrated by it.

One reading explicitly left open

The interpretation of pointed-base goblets as single-use volumetric gauges is not settled — they have also been read as disposable drinking or ritual cups — and is advanced as a hypothesis consistent with the metrological model, not as proof of it.

// access and citation

Open access, permanent identifier

Diener, Ł. (2026). Excel in Clay II: The Meluhha Operating System — A Mathematical Reconstruction of the Indus Valley Non-Linguistic Cargo-Tag Database (version 1.0). Zenodo. https://doi.org/10.5281/zenodo.20688721

Preprint, not peer reviewed by a journal. Licensed CC BY 4.0. Second of three audits applying the projection-before-meaning protocol to undeciphered Bronze Age administrative corpora.

Read on Zenodo Excel in Clay I Excel in Clay III All six audits