BitConcepts · tbitcs · May 22, 2026
diff --git a/research/indus/indus-anchor-model/CITATION.cff b/research/indus/indus-anchor-model/CITATION.cff
@@ -0,0 +1,42 @@
+cff-version: 1.2.0
+message: "If you use this software or the research outputs in this repository, please cite both the preprint and the software as below."
+type: software
+title: "Glossa-Lab: AI-Assisted Computational Linguistics Research Platform"
+version: "1.0.0"
+date-released: "2026-05-21"
+license: MIT
+repository-code: "https://github.com/BitConcepts/glossa-lab"
+url: "https://github.com/BitConcepts/glossa-lab"
+authors:
+  - family-names: Pierson
+    given-names: Tristen Kyle
+    affiliation: "BitConcepts LLC"
+    email: tpierson@bitconcepts.tech
+
+abstract: >
+  Glossa-Lab is a computational linguistics research platform for statistical
+  analysis, decipherment, and hypothesis testing of ancient and undeciphered
+  writing systems. It was used to produce the Indus Script decipherment preprint
+  "A Falsifiable Computational Decipherment Hypothesis for the Indus Valley Script"
+  (Pierson 2026).
+
+keywords:
+  - computational linguistics
+  - Indus Valley Script
+  - decipherment
+  - ancient scripts
+  - proto-Dravidian
+  - corpus analysis
+
+references:
+  - type: article
+    title: >
+      A Falsifiable Computational Decipherment Hypothesis for the Indus Valley Script:
+      161 Candidate Proto-Dravidian Anchors and a Three-Slot Positional Grammar
+    authors:
+      - family-names: Pierson
+        given-names: Tristen Kyle
+        affiliation: "BitConcepts LLC"
+    year: 2026
+    notes: "Preprint v1 — Not peer-reviewed"
+    url: "https://github.com/BitConcepts/glossa-lab/tree/main/research/indus"
diff --git a/research/indus/indus-anchor-model/README.md b/research/indus/indus-anchor-model/README.md
@@ -0,0 +1,167 @@
+# A Falsifiable Computational Anchor Model for the Indus Script
+## Preprint Release v1 — Reproducibility Package
+
+**Pierson, T. K. (2026).** *A Falsifiable Computational Decipherment Hypothesis for the Indus Valley Script: 161 Candidate Proto-Dravidian Anchors and a Three-Slot Positional Grammar.* Preprint v1.
+
+> This repository accompanies a preprint. It does not claim final decipherment. It provides data tables, scripts, and validation outputs for testing a candidate positional-grammar and anchor-reading model. Full reproduction of corpus-level counts may require access to the Holdat LLC Indus Corpus v3 or rerunning the pipeline on a compatible public corpus such as ICIT after sign-code crosswalking.
+
+---
+
+## 1. What is this repository?
+
+This package provides the public materials for a computational study of Indus-script structure using positional analysis, bigram collocations, permutation tests, fish-sign polysemy testing, betweenness centrality stratification, and candidate sign-anchor modeling.
+
+**The study does not claim final decipherment.** It proposes a falsifiable model with explicit confidence tiers, documented limitations, and cross-corpus validation targets. The primary corpus (Holdat LLC Indus Corpus v3, Miller 2025) is not redistributed here due to licensing.
+
+---
+
+## 2. What claims can be reproduced from public data?
+
+The following can be tested using only the public tables in `data/public/`:
+
+| Claim | Script | Status |
+|---|---|---|
+| Fish signs are 0% isolated in all corpus contexts | `scripts/fish_sign_test/` | REPRODUCED_FROM_PUBLIC_DATA |
+| M342·M176 is the highest-PMI bigram pair | `scripts/grammar_model/` | REPRODUCED_FROM_PUBLIC_DATA |
+| 161 H+M signs have BC > 0 (grammar candidates) | `scripts/network_analysis/` | REPRODUCED_FROM_PUBLIC_DATA |
+| Anchor table covers 397 Mahadevan sign numbers | integrity check | REPRODUCED_FROM_PUBLIC_DATA |
+| 141/161 H+M signs have BC = 0 (name-syllable signature) | `scripts/network_analysis/` | REPRODUCED_FROM_PUBLIC_DATA |
+
+Run all public checks at once:
+
+```bash
+cd scripts/validation/
+python run_all_public_checks.py
+# Output: outputs/logs/public_validation_report.txt
+```
+
+---
+
+## 3. What cannot be reproduced without restricted corpus access?
+
+The following require authorized access to the Holdat LLC Indus Corpus v3 or an equivalent corpus (e.g. ICIT after sign-code crosswalking):
+
+- Token coverage percentages (90.96% H+M, 9.34% LOW)
+- Positional rates (I/M/T percentages) from raw seal sequences
+- Permutation null test (z = 10.3, 0/2000 permutations)
+- Bootstrap confidence intervals for site-level KL divergences
+- Site-level sign frequency tables beyond the 5-site summary
+
+See `data/restricted/README_restricted_data.md`.
+
+---
+
+## 4. What data are public?
+
+All files in `data/public/` are released under CC BY 4.0:
+
+| File | Description |
+|---|---|
+| `anchor_table_397.csv` | All 397 Mahadevan M-number signs with candidate readings and confidence levels |
+| `fish_sign_contexts.csv` | Per-seal compound-context listing for M047 (plain fish) and M001 |
+| `formula_bigrams.csv` | Top-30 H+M×H+M bigrams with counts and PMI |
+| `iconographic_formula_pairs.csv` | 63 enriched INITIAL-sign × seal-icon chi-square pairs |
+| `polysemy_divergence_summary.csv` | Phase-150 permutation null test (21 signs, 1,000 shuffles) |
+
+---
+
+## 5. What data are restricted?
+
+The Holdat LLC Indus Corpus v3 (Miller 2025) is not redistributed here. See `data/restricted/README_restricted_data.md` for details on what corpus-level outputs require restricted access and how to request it.
+
+---
+
+## 6. How do I run each analysis?
+
+Each script folder contains a `README.md` with specific instructions. Quick start:
+
+```bash
+# Install dependencies
+pip install -r requirements.txt
+
+# Run all public reproducibility checks
+python scripts/validation/run_all_public_checks.py \
+    --data-dir data/public/ \
+    --output-dir outputs/
+
+# Run betweenness centrality on public bigram table
+python scripts/network_analysis/run_betweenness.py \
+    --bigrams-file data/public/formula_bigrams.csv \
+    --hm-signs-file data/public/anchor_table_397.csv \
+    --output-dir outputs/
+
+# Run fish-sign isolation test
+python scripts/fish_sign_test/run_fish_sign_test.py \
+    --contexts-file data/public/fish_sign_contexts.csv \
+    --output-dir outputs/
+```
+
+---
+
+## 7. What outputs should I expect?
+
+After running the public validation suite:
+
+```
+outputs/logs/public_validation_report.txt   — human-readable check results
+outputs/tables/public_validation_summary.csv — machine-readable status per check
+outputs/tables/bc_rankings.csv              — betweenness centrality rankings
+outputs/tables/fish_sign_summary.csv        — fish-sign isolation test results
+```
+
+Each output file notes whether it was produced from public data alone or requires restricted corpus access.
+
+---
+
+## 8. How do I cite this work?
+
+```bibtex
+@misc{pierson2026indus,
+  author    = {Pierson, Tristen Kyle},
+  title     = {A Falsifiable Computational Decipherment Hypothesis for the
+               Indus Valley Script: 161 Candidate Proto-Dravidian Anchors
+               and a Three-Slot Positional Grammar},
+  year      = {2026},
+  note      = {Preprint v1 -- Not peer-reviewed},
+  url       = {https://github.com/BitConcepts/glossa-lab}
+}
+```
+
+See `CITATION.cff` for CFF 1.2.0 format.
+
+---
+
+## 9. What would falsify the model?
+
+The model makes falsifiable predictions. It is weakened or refuted if:
+
+**Structural falsification (Tier 1):**
+- Fish signs appear in isolation in formal seal contexts at >0% rate
+- Positional structure is indistinguishable from random in a permutation test on ICIT
+- The three-slot grammar fails to predict sign position at better than chance on a held-out corpus
+- Site-level KL divergences are not bootstrap-robust on ICIT
+
+**Candidate anchor falsification (Tier 2):**
+- H+M token coverage collapses to <70% under ICIT sign crosswalk
+- Proposed terminal signs do not remain terminal-dominant in ICIT positional profiles
+- M267-equivalent sign behaves iconographically rather than functionally in ICIT
+- Bigram/centrality structure does not reproduce on ICIT
+
+**External anchor falsification:**
+- The Shu-ilishu phonological test fails at >3/4 slots on a larger corpus
+- The M267 correction is contradicted by newly discovered bilingual inscription evidence
+
+See `docs/falsification_protocol.md` for the complete formal checklist.
+
+---
+
+## License
+
+- Code and scripts: MIT License
+- Documentation and public tables: CC BY 4.0
+
+The Holdat LLC Indus Corpus v3 is not redistributed. See `data/restricted/README_restricted_data.md`.
+
+## Repository
+
+https://github.com/BitConcepts/glossa-lab