Indus-Valley

Package Install

Installable package: python3.11 -m pip install gnosis-indus. Current release: 0.1.0 on PyPI. Source: Zer0pa/Indus-Valley.

python3.11 -m pip install gnosis-indus

For full install, smoke, source, and developer commands, click here.

---

00 · GNOSIS-INDUS-VALLEY · NON-DECIPHERMENT SEARCH RESEARCH-READY · v0.1.0 4,500 Years Undeciphered, Now Searchable Live experiment. Not a release. Ambition headlined. Claims bounded. Gnosis Indus-Valley · PyPI gnosis-indus v0.1.0 · 412 signs · 70 clusters · github.com/Zer0pa/Indus-Valley The Indus script has gone unread for a century. Thousands of marks on seals, tablets, and tools — catalogued, contested, photographed — and no one knows what they say. What was missing was not more scholarship. It was a way to search the corpus by the shape of the marks, without claiming what they mean. gnosis-indus ships a clean-room 412-sign / 70-cluster catalogue and answers shape queries in 0.0451 ms. No glyph is read.

Scope: 412-sign clean-room catalogue and 70 shape clusters. Query speed is measured; no glyph is read.

01 · THE GAP A CENTURY WITHOUT SHAPE SEARCH The Indus script has gone undecoded for a century; its marks were never searchable by shape.

02 · MARKETS ADJACENT FORECASTS Cultural heritage digitization '30 · $8.1B Research data management '30 · $6.7B Scholarly infrastructure '30 · $5.3B Digital humanities '30 · $3.2B AI for archaeology '30 · $1.4B source: heritage and research-infrastructure forecasts. Best-fit users: epigraphy labs, museum archives, and digital-humanities groups. No traction or TAM claim.

03 · VALUE OF MARKET $8.1 B Heritage digitization is funded. Tools that index this script by shape, without claiming to read it, remain almost absent.

04 · INSIGHT The mark has a shape. The shape can be searched.

05.1 · CURRENT TECH CATALOGUED, NOT SEARCHABLE BY FORM Indus scholarship lives in sign catalogues, visual classifications, and concordances. They let scholars compare marks one at a time, but no tool let them query the whole corpus by the shape of a single sign.

05.2 · OUR TECH SEARCH BY SHAPE gnosis-indus is a clean-room runtime: no neural model, no learned embedding, no substrate hypothesis. It bundles a conditional 412-sign / 70-cluster catalogue and answers shape queries in 0.0451 ms. The full 179-inscription corpus stays fetch-external under data policy. No glyph is read; only the shape is found.

05.3 · BENCHMARKS SEARCH TARGET + NMI Query latency0.0451 ms target100 ms NMI0.5793vs ICIT sigma5.65 Search PASS NMI PASS Substrate OPEN Result: demo shape search 0.0451 ms vs 100 ms target · NMI 0.5793 vs ICIT Sets.

06 · MEASUREMENT SEARCH TARGET · NMI · STRUCTURE Shape search clears the latency target. Decipherment is not measured — deliberately.

06.1 · COMPARATIVE PERFORMANCE · DEMO FIXTURE Max query 0.0451 ms target 100 ms under ~2,217× NMI 0.5793 sigma 5.65 Demo shape search at 0.0451 ms against a 100 ms target — roughly 2,217× under. Catalogue NMI 0.5793 vs ICIT Sets at sigma 5.65. Phase 5 substrate question unresolved; no reading claimed.

07 · KEY METRICS MEASURED RESULTS

07.1 · MAX QUERY LATENCY 0.0451ms Demo shape search · vs 100 ms target

07.2 · CATALOGUE NMI 0.5793 vs ICIT Sets · sigma 5.65

07.3 · PHASE 5 VERDICT STRUCT Linguistic structure confirmed · substrate unresolved

07.4 · PYPI RELEASE 0.1.0 PyPI gnosis-indus · public, reproducible

07.5 · CATALOGUE SCOPE 412signs 70 clusters · 179 inscriptions; full corpus fetched separately

08 · DETERMINISM SEARCH THAT REPLAYS Same query. Same fixture. Same matches — across replay clones.

08.1 · WHAT REPLAYS EXACTLY DEMO SEARCH SURFACE The runtime is clean-room — no neural model, no learned embeddings, no substrate guess. The same shape query against the bundled fixture returns the same matches across replay clones, every time, anchored to the same source records. The catalogue is conditional on Phase 4 stability, and the Phase 5 source-structure question is unresolved. Determinism covers shape search, not interpretation. No glyph is read, and the runtime is built so it cannot quietly start to.

08.2 · HONEST BLOCKER Honest Blocker · No reading claimed. The Phase 5 source-structure question is unresolved: we do not know what language, if any, the script encodes. Sign images stay policy-limited under DATA_POLICY.md. The bundled fixture is small; the full 412 / 70 / 179 catalogue is fetched separately.

09 SEARCH THAT REFUSES TO READ.

09.1 · THE AMBITION The ambition is computational restraint. gnosis-indus gives Indus scholarship a search surface, a cluster map, and a comparison primitive — and stops there. The refusal to read is not a hedge. It is the product. A catalogue researchers and museums can trust because it never overreaches into translation.

09.2 · WHAT WORKS NOW Shape search runs in 0.0451 ms, NMI confirms cluster structure, and the catalogue stays free of reading.

09.3 · WHAT'S STILL OPEN Substrate question unresolved, image policy limited, catalogue conditional, and full corpus stays fetch-external for now.

09.4 · SCHOLARSHIP · NEAR-TERM (12–24 MO) Indus scholars can search by shape A paleographer who notices a familiar mark on a newly photographed seal can find every visually related sign across the corpus in milliseconds. The conversation moves from "have I seen this before" to "here are the seventeen places it appears."

09.5 · DEBATE · NEAR-TERM (12–24 MO) Decipherment debate gets cleaner footing Comparisons among signs and clusters can stay anchored to structure, not to translation guesses. Researchers arguing competing theories share the same shape-grounded reference instead of talking past each other about disputed readings.

09.6 · MUSEUMS · MID-TERM (24–48 MO) Museum catalogues gain a structural index Curators can attach sign-family and cluster labels to seal records alongside images. Visitors and remote researchers query holdings by mark shape, and discovery surfaces stop depending on a translation no museum is willing to assert.

09.7 · METHOD · MID-TERM (24–48 MO) Shape search becomes shared infrastructure The same primitive — search by form, refuse to read — generalises to other undeciphered scripts: Linear A, Proto-Elamite, Rongorongo. Shape-first analysis becomes a shared method, and labs stop rebuilding the same catalogue plumbing from scratch.

09.8 · DISCIPLINE · PARADIGM (48 MO+) Honest non-claim becomes a scientific asset In contested heritage, a maintained refusal to overclaim outlasts every dramatic weak claim. The Indus catalogue stands as the durable object: a corpus researchers, curators, and the public can trust precisely because it never said what the signs mean.

---

Install / Developer Commands Detailed

Package Install

Installable package: python3.11 -m pip install gnosis-indus. Current release: 0.1.0 on PyPI. Source: Zer0pa/Indus-Valley.

python3.11 -m pip install gnosis-indus

Import smoke:

python3.11 - <<'PY'
import importlib.metadata as md
import gnosis_indus

print("gnosis-indus", md.version("gnosis-indus"))
PY

Install success only proves package acquisition/import. Product scope, stale PyPI state, platform limits, and blockers remain in the front-door sections below.

Quick Start

Reproduce the Phase 02 stronger smoke path on any clean Python 3.11 host:

git clone https://github.com/Zer0pa/Indus-Valley.git gnosis-indus
cd gnosis-indus
python3.11 -m venv .venv && source .venv/bin/activate
pip install -e ".[test,numerics]"
pytest -q

Expected: 14 passed. The pytest suite reproduces the authority-doc query records from authority/review_pack/search_demo_summary.md against the bundled artifacts/phase4/indus_catalogue_demo_fixture.json. The fixture is small and authority-anchored; the real full catalogue stays FETCH_EXTERNAL per DATA_POLICY.md. The Phase 4 stability caveat (k=70 conditional) remains visible in the package and fixture surfaces.