Gnosis Glyph Engine

This is a live window into the Zer0pa lab. None of the work in this repo is a final release. Zer0pa GitHub repos are open as a matter of course for visibility and community participation. Always-in-beta is the posture: useful now, improving continuously. Open gates and honest blockers are surfaced in plain text below — see §5 "What We Don't Claim" and §11 "Upcoming Workstreams".

Licensing

This repository is part of the Zer0pa Gnosis Portfolio.

Code in this repository is licensed under the Apache License 2.0. See LICENSE for the full text. SPDX identifier: Apache-2.0.

Documentation, reports, and written materials are licensed under Creative Commons Attribution 4.0 International. SPDX identifier: CC-BY-4.0. Canonical terms: https://creativecommons.org/licenses/by/4.0/.

Data and fixtures are handled per dataset and artifact family. See DATA_POLICY.md for this repository's data boundary. The code license does not license raw corpora, image-bearing cultural-heritage assets, private HF artifacts, model weights, endpoint logs, or operational transcripts.

Trademarks - "Gnosis", "Zer0pa Gnosis", and distinctive sub-marks are trademarks of Zer0pa. Apache-2.0 and CC-BY-4.0 do not grant trademark rights. See TRADEMARKS.md.

What This Is

gnosis-glyph-engine is the Gnosis research lane that asks one falsifiable question:

Is a standalone glyph-engine package boundary earned by a real consumer, a monorepo-free smoke path, and a descriptor ablation against borrowed baselines?

It is an extraction-target scaffold for reusable glyph geometry, stroke, and descriptor kernels. It is not a public library, not a product, and not a decipherment claim.

Defensible headline metrics (as of 2026-04-28): Borrowed-baseline ablation across 10 seeds: mean σ baseline_orb 4.14 ± 1.12, baseline_hu_regionprops 2.95 ± 1.06, baseline_hog 1.15 ± 1.17 (artifact: artifacts/robustness/robustness_report.json). Determinism: replay_all_identical == true for every arm in the seed-42 ablation; reference-freeze SHA256 is byte-stable across arms. 17 pytest tests pass when sibling gnosis-morph-bench is installed; 16 pass + 1 cleanly skips when it is not — preserving repo independence.

Honest blocker: The owned-arm authority metric package_boundary_earned is UNTESTED. Two Indus source files (scripts/indus/stroke_native_encoding.py, scripts/indus/phase3_common.py) are not present in the portfolio snapshot; D-06 retrieval is the unblock. Phase 02b cannot start until D-06 clears.

What We Prove

• The frozen Phase 01 interface (Descriptor, LearnedDescriptor, manifest_builder) is implementable using only borrowed OSS — OpenCV, scikit-image, scikit-learn, NumPy.
• The package installs cleanly with pip install -e .[dev] on Python 3.13 with no live-monorepo imports and no monorepo path coupling.
• The 17-test pytest suite passes when sibling gnosis-morph-bench is installed; 16 pass + 1 cleanly skips when it is not — preserving repo independence.
• The frozen BenchmarkManifest shape is consumed unmodified by gnosis_morph_bench.schema.load_manifest, producing finite sigma, null_mean, null_std, silhouette, and mean_jaccard per arm.
• A multi-seed (10-seed) borrowed-baseline ceiling is recorded in artifacts/robustness/robustness_report.json; mean σ across seeds comes out to baseline_orb 4.14 ± 1.12, baseline_hu_regionprops 2.95 ± 1.06, baseline_hog 1.15 ± 1.17.
• Determinism is checked: replay_all_identical == true for every arm in the seed-42 ablation; reference-freeze SHA256 is byte-stable across arms.

What We Don't Claim

• We do not claim that any owned descriptor beats borrowed baselines. Phase 02b is BLOCKED on D-06 (two Indus source files are not present in the portfolio snapshot) and the gate-4 verdict is UNTESTED.
• We do not claim cross-corpus generalisation. The fixture is a deterministic 12-glyph synthetic generator; rights-cleared real-glyph evidence does not yet exist in this repo.
• We do not claim public-release readiness, performance superiority, or product readiness.
• We do not claim Indus or cuneiform domain results — those are owned by sibling lanes.

Tests and Verification

• Local CI: .github/workflows/ci.yml runs pip install -e .[dev] then pytest -q on every push and PR to main.

• Local repro:

  python3.13 -m venv .venv && source .venv/bin/activate
  pip install -e .[dev]
  pytest -q                         # → 16 passed, 1 skipped
  pip install -e /path/to/gnosis-morph-bench    # optional contract test
  pytest -q                         # → 17 passed
  python -m gnosis_glyph_engine.scripts.run_ablation
  python -m gnosis_glyph_engine.scripts.run_robustness

• Provenance: seed=42 ablation matches .gpd/phases/02-minimal-extraction-smoke/02-01-VERIFICATION.md to 6 decimal places; seeds=0..9 robustness matches 02-02-VERIFICATION.md.

• Endpoint-leak scan is part of the closeout verification (Gate D in 02-02-VERIFICATION.md); see docs/PROVENANCE_LABELS.md for the symbolic-label convention.

Proof Anchors

Claim	Evidence
Authority metric definition	SOVEREIGN_PRD.md §Authority Metric
Frozen consumer + interface	.gpd/phases/01-consumer-and-interface-freeze/01-DECISIONS.md D-01, D-02
Frozen fixture + ablation rule	.gpd/phases/01-consumer-and-interface-freeze/01-DECISIONS.md D-03, D-04
Phase 02a single-seed ablation	artifacts/ablation/ablation_report.json + .gpd/phases/02-minimal-extraction-smoke/02-01-VERIFICATION.md
Phase 02c multi-seed robustness	artifacts/robustness/robustness_report.json + .gpd/phases/02-minimal-extraction-smoke/02-02-VERIFICATION.md
Sovereign-vs-adapter scope decision	.gpd/phases/02-minimal-extraction-smoke/02-02-DECISIONS.md D-02c-02
Owned-arm source-retrieval blocker	SOURCE_BOUNDARY.md (D-06)
HF custody truth	HF_CUSTODY_REGISTER.md (canonical: Architect-Prime/glyph-engine-artefacts; Zer0pa namespace empty for this lane post-2026-04-27 migration)
Path/endpoint scrub convention	docs/PROVENANCE_LABELS.md
Auditor fast path	AUDITOR_PLAYBOOK.md

Commercial Readiness

Field	Value
Verdict	BLOCKED
Posture	package_boundary_unearned_pending_owned_arm_d06

The authority metric package_boundary_earned is UNTESTED. Gate 4 (owned arm ablation) cannot run until D-06 clears. The Verdict cell draws from the locked enum {STAGED, PASS, PARTIAL, BLOCKED, FAIL, INCONCLUSIVE}.

No Confidence row: no artifact backs a confidence estimate for Phase 03 readiness.

Repo Shape

.
├── NOTICE                                  # Apache-2.0 code and CC-BY-4.0 docs posture
├── HF_CUSTODY_REGISTER.md                  # HF artefact-dataset truth
├── README.md                               # this file
├── SOVEREIGN_PRD.md                        # authority metric + extraction gate
├── SOURCE_BOUNDARY.md                      # candidate sources, ledger, D-06
├── AUDITOR_PLAYBOOK.md                     # fast path + reproduce + claim-replay
├── CONTRIBUTING.md, GOVERNANCE.md, RELEASING.md, ROADMAP.md, SECURITY.md
├── CHANGELOG.md, CITATION.cff, CODE_OF_CONDUCT.md
├── DATA_POLICY.md, PUBLIC_AUDIT_LIMITS.md
├── THIRD_PARTY_NOTICES.md, TRADEMARKS.md
├── pyproject.toml                          # gnosis-glyph-engine==0.1.0a1
├── _internal/                              # agent-orchestration scaffolding (not reader-facing)
│   ├── AUTONOMOUS_EXECUTION_POLICY.md
│   ├── GPD_BOOTSTRAP_GUIDE.md
│   ├── UNIVERSAL_STARTUP_PROMPT.md
│   ├── TEMPLATE_USAGE.md
│   ├── MIGRATION_PLAN.md
│   ├── WORKSTREAM_GPD_INIT_CHECKLIST.md
│   ├── TODO.md
│   └── AGENT_STATUS_REPORT_2026-04-24.md
├── .github/
│   ├── workflows/ci.yml
│   ├── ISSUE_TEMPLATE/
│   └── PULL_REQUEST_TEMPLATE.md
├── .gpd/
│   ├── PROJECT.md, STATE.md, ROADMAP.md, REQUIREMENTS.md, DECISIONS.md,
│   │   CONVENTIONS.md, NOTATION_GLOSSARY.md, state.json, config.json
│   └── phases/
│       ├── 00-workstream-bootstrap/*
│       ├── 01-consumer-and-interface-freeze/*
│       └── 02-minimal-extraction-smoke/*      # 02a + 02c artefacts
├── src/gnosis_glyph_engine/
│   ├── __init__.py, protocols.py, fixtures.py, manifest_builder.py
│   ├── baselines/{__init__,orb,hu_regionprops,hog}.py
│   ├── owned/__init__.py                      # raises SourceRetrievalPending until D-06
│   └── scripts/{__init__,run_ablation,run_robustness}.py
├── tests/{test_fixtures,test_baselines,test_manifest_builder}.py
├── scripts/run_ablation.py                    # thin shim
├── artifacts/
│   ├── ablation/   {ablation_report.json, per_arm/baseline_*.{manifest,smoke_report}.json}
│   └── robustness/ robustness_report.json
├── code/README.md                             # legacy code-surface marker
└── docs/{ARCHITECTURE,LEGAL_BOUNDARIES,PROVENANCE_LABELS,FAQ,SUPPORT,README}.md

Quick Start

git clone https://github.com/Zer0pa/Glyph-Engine.git
cd Glyph-Engine
python3.13 -m venv .venv && source .venv/bin/activate
pip install -e .[dev]
pytest -q                          # → 16 passed, 1 skipped
python -m gnosis_glyph_engine.scripts.run_ablation
python -m gnosis_glyph_engine.scripts.run_robustness

Console scripts (after install): glyph-engine-ablation, glyph-engine-robustness.

Current Gaps

• D-06 (owner action): retrieve scripts/indus/stroke_native_encoding.py and scripts/indus/phase3_common.py from the live monorepo, or grant pod access to a live-monorepo clone. Phase 02b cannot start until D-06 clears.
• No real-glyph fixture; the synthetic 12-glyph generator is the only ablation surface today. Rights-cleared real-glyph evidence does not yet exist in this repo.
• Public contact surface remains owner-held.

§11 Upcoming Workstreams

Operations / External Dependency

• D-06 source-file retrieval — owner action: retrieve scripts/indus/stroke_native_encoding.py and scripts/indus/phase3_common.py from the live monorepo, or grant pod access to a monorepo clone. All Phase 02b work is gated on this.

Research-Deferred — Investigation Underway

• Real-glyph fixture — replace the deterministic 12-glyph synthetic generator with rights-cleared real-glyph evidence. No rights-cleared fixture exists yet; investigation of sourcing options is underway. Cross- corpus generalisation claims cannot be made until this is resolved.

Active Engineering

• Phase 02b — owned-arm gate testing — implement src/gnosis_glyph_engine/owned/stroke_compass.py and src/gnosis_glyph_engine/owned/topology.py from the retrieved Indus source files; run ablation + robustness against the Phase 02c baseline ceiling; apply D-04 pass rule under D-02c-02 sovereign routing. Starts the moment D-06 clears.

Zero-Base Scientific Thinking — GPD Research and Planning Pending

• None at this time.