Metrological Domain Profiling (MDP)

A Computational Method for Classifying Signs in Undeciphered Administrative Scripts

Author: Adrian Sharman, SoulDriver Research (souldriver.com.au)

Paper: Metrological Domain Profiling: A Computational Approach to Administrative Architectures in Proto-Cuneiform, Proto-Elamite, and the Indus Valley Script (Sharman, April 2026)

---

Overview

This repository contains the complete code and methodology for Metrological Domain Profiling (MDP), a novel computational method that classifies signs in ancient administrative scripts by analysing their statistical co-occurrence with metrological numeral systems. The method requires no linguistic assumptions, no phonetic hypotheses, and no prior knowledge of sign meanings.

Key Results

Script	Period	Corpus Size	Method	Accuracy / Finding
Proto-Cuneiform	c. 3350–3000 BCE	31,191 signs, 6,819 tablets	Subtractive MDP	100% accuracy on 28-sign validation set
Proto-Elamite	c. 3100–2900 BCE	14,716 records	Standard MDP	79 signs classified, 64% corpus coverage
Indus Valley	c. 2600–1900 BCE	11,280 signs, 2,543 inscriptions	Animal-Sign Enrichment	Semantic locks discovered (G321→Hare: 195.6× enrichment)

---

Repository Contents

File	Description
protocuneiform_mdp.py	Study 1: Validates Subtractive MDP on proto-cuneiform (Uruk IV/III). Achieves 100% on 28 known signs.
protoelamite_mdp.py	Study 2: Applies MDP to Proto-Elamite. Classifies 79 signs, reveals syntax and seal specialisation.
indus_mdp.py	Study 3: Adapts methodology for the Indus Valley Script. Discovers animal-sign semantic locks.
Indus_scripts_terminal_output.md	Full terminal output from the Indus analysis run.
README.md	This file.

---

Data Sources & Reproduction Instructions

This repository does not include the source datasets due to size constraints. All data is freely available from the following sources:

1. CDLI Bulk Data (Proto-Cuneiform & Proto-Elamite)

Both protocuneiform_mdp.py and protoelamite_mdp.py require two files from the Cuneiform Digital Library Initiative (CDLI):

File	Description	Approx. Size
cdliatf_unblocked.atf	Bulk ATF transliteration dump — contains transliterations of ~135,000 cuneiform tablets	~83 MB
cdli_cat.csv	Catalogue with metadata (period, provenience, language) for each tablet	~148 MB

How to obtain:

1. Visit https://cdli.earth/ (or https://github.com/cdli-gh/data)
2. Download the bulk ATF transliteration file and the catalogue CSV
3. Place both files in a local directory (e.g., C:\Users\you\cdli-data\)
4. Update the ATF_FILE and CAT_FILE path variables at the top of each script

The CDLI data is open-access and freely available for research purposes.

2. Yajnadevam Indus Corpus (Indus Valley Script)

indus_mdp.py requires the SQL population script from the yajnadevam/indus-website repository, which contains a structured digital extraction of the Interactive Corpus of Indus Texts (ICIT; Wells & Fuls 2015).

File	Description	Approx. Size
population-script.sql	MySQL dump containing 2,543 inscriptions with sign sequences, iconographic descriptions, and site provenance	~430 KB

How to obtain:

1. Clone or download https://github.com/yajnadevam/indus-website
2. The file population-script.sql is in the repository root
3. Update the SQL_FILE path variable at the top of indus_mdp.py

Note: The script parses the SQL file directly using regex — no MySQL installation is required.

3. Alternative Indus Corpus (Smaller, JSON format)

For a smaller Indus corpus (179 Mohenjo-daro unicorn seals in JSON format), see https://github.com/mayig/indus-valley-script-corpus. This corpus was used for initial testing but lacks the multi-site, multi-animal coverage needed for the semantic lock analysis.

---

Requirements

• Python 3.8+
• No external dependencies — uses only re, csv, os, json, and collections from the Python standard library
• Each script runs in under 60 seconds on a standard laptop

Usage

# 1. Download the data sources listed above
# 2. Update file paths in each script to point to your local data files
# 3. Run:

python protocuneiform_mdp.py    # Study 1: Proto-Cuneiform Validation (expect: 28/28 = 100%)
python protoelamite_mdp.py      # Study 2: Proto-Elamite Application
python indus_mdp.py             # Study 3: Indus Valley Discovery

---

Methodology: Subtractive Metrological Domain Profiling

Ancient administrative scripts pair commodity signs with numeral notations drawn from specific metrological systems. Grain is measured in capacity units (with fractional subdivisions), livestock is counted in discrete integers, and land is measured in area units. MDP exploits these metrological signatures to classify signs by commodity domain.

The Polyvalency Problem

Proto-cuneiform uses polyvalent numeral signs — the same grapheme (N14) represents different values in different metrological systems. As Englund (2011) describes: "the sign N14 can represent ten clay pots of butter oil, a measure of grain corresponding to about 150 litres of barley, or a field of about 6 hectares." Standard numeral co-occurrence analysis fails because the numeral codes are domain-ambiguous.

Proto-Elamite, by contrast, evolved dedicated numeral codes per domain (N39B exclusively for grain capacity, N23/N51 exclusively for the decimal/animate system). This architectural difference — shared vs dedicated numeral codes — is itself a finding about how the two contemporary writing systems diverged in encoding metrological information.

The Subtractive Solution

Subtractive MDP overcomes the polyvalency barrier with a three-step sieve:

1. Total Numeral Adjacency: Is the sign counted at all? (Uses all numerals including polyvalent N01/N14/N34)
2. System-Specific Anchor Check: Does it co-occur with fractional anchors unique to a specific system?
   • N39A, N24 → Grain capacity (ŠE system)
   • N51, N54 → Bisexagesimal rations
   • N50, N47, N08 → Area/land (GAN₂ system)
3. Subtractive Deduction: If counted but no system-specific anchors → sexagesimal discrete (livestock, textiles, metals, people)

Development Trajectory

The final 100% accuracy was achieved through iterative refinement:

Version	Accuracy	Key Change
v1	16% (3/19)	Hard-coded Proto-Elamite numeral systems — wrong for proto-cuneiform
v2	22% (4/18)	Data-driven anchor discovery — no livestock discriminators found
v3	55% (6/11)	Subtractive method introduced — grain failed due to anchor rarity
v4	94% (15/16)	Data normalisation fixes (tilde stripping, lowercase purge, compounds)
v4 expanded	100% (28/28)	Corrected validation dictionary + expanded test set

---

The Indus Discovery

By adapting MDP to analyse sign–animal co-occurrence instead of sign–numeral co-occurrence, we discovered that specific Indus signs are statistically locked to specific seal animals:

Sign	Animal	Enrichment	Exclusivity
G321	Hare	195.6×	12/12 (100%)
G850	Anthropomorphic	60.5×	10/10 (100%)
G845	Hare	63.6×	Shared with Anthropomorphic
G407	Hare	50.9×	Primary hare, shared
G48	Elephant	35.6×	Near-exclusive
G318	Rhinoceros	37.0×	Near-exclusive
G436	Rhinoceros	31.7×	Near-exclusive
G923	Elephant	14.8×	Exclusive

Each animal type possesses a unique sign vocabulary — a set of enriched signs that collectively distinguish its inscriptions from all others. This provides the first corpus-scale statistical evidence that Indus seal animals encode institutional/departmental identity, paralleling the commodity-specific seal ownership discovered in Proto-Elamite.

---

Research Infrastructure

This research was conducted using KARP (Knowledge Acquisition Research Protocol), a multi-agent AI deliberation system developed by SoulDriver Research that orchestrates Claude (Anthropic), GPT, Gemini (Google), and Grok through structured adversarial research sessions.

The initial hypothesis — that Proto-Elamite's dedicated numeral codes could serve as a computational anchor for unsupervised sign classification — emerged from KARP council sessions on undeciphered ancient scripts. The Subtractive MDP variant was proposed during a Gemini review that identified the polyvalency barrier. Three critical data-normalisation fixes were likewise identified through external AI review.

The complete research trajectory — from initial hypothesis through five algorithm iterations to the final 100% validation, plus the Indus semantic lock discovery — was conducted in a single research session, demonstrating the potential of structured multi-agent AI deliberation for accelerating computational humanities research.

For more on KARP and SoulDriver Research, visit souldriver.com.au.

---

License

MIT License — see LICENSE for details.

Citation

If you use this code or methodology, please cite:

Sharman, A. (2026). Metrological Domain Profiling: A Computational Approach to
Administrative Architectures in Proto-Cuneiform, Proto-Elamite, and the Indus Valley
Script. SoulDriver Research. https://souldriver.com.au

Acknowledgements

• CDLI (Cuneiform Digital Library Initiative) for open-access cuneiform data
• Bryan K. Wells and Andreas Fuls for the Interactive Corpus of Indus Texts (ICIT)
• yajnadevam for the digital ICIT extraction
• mayig for the Indus Valley script corpus (JSON format)
• Robert K. Englund and Peter Damerow for the foundational work on proto-cuneiform numeral systems
• Jacob L. Dahl for Proto-Elamite sign identifications