Gnosis Cuneiform

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 ## What We Don't Claim and ## 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.

Public visibility is a separate repository-setting action. The license files in this repo define the intended open-source/open-documentation terms for released Gnosis code and written materials; they do not publish rights-gated data.

Traditional-Knowledge Acknowledgment

This repository works with material in cuneiform-bearing writing systems (Sumerian, Akkadian, Babylonian, and related). Zer0pa claims no proprietary right over cuneiform, its scripts, or the archaeological record.

This repository preserves a negative-control result as a first-class output; it does not claim a definitive decipherment of cuneiform or related scripts.

Good-faith inquiries from identified communities, governmental antiquity authorities, or institutional bodies may be sent to architects@zer0pa.ai.

What This Is

gnosis-cuneiform is a negative-control / control-pack repository inside the Gnosis methods portfolio. It preserves the cuneiform benchmark/control artefact chain, the failed governing gate, the source boundary, and the rerun contract without pretending the lane is a public product or a successful scientific flagship. It ships a stdlib-only manifest-validation smoke surface so the custody and shape of pinned inherited artefacts are mechanically checkable from a fresh checkout.

Headline metric: pytest -q → 3 passed in 0.04s; 5/5 cross-field invariants hold on real-manifest smoke (artefacts/smoke/manifest_validation_report.json); 6/6 SHA-256 pinned upstream manifest custody verified against Architect-Prime/cuneiform-control-artefacts (token-verified 2026-04-27, docs/HF_CUSTODY_REGISTER.md Verification 5).

Honest blocker: The inherited governing gate NO_GO_GOVERNING_GATE_UNMET failed at governing_1nn_accuracy = 0.021916 and remains failed. This lane preserves the negative result as a first-class output; repairing the gate is out of scope for this lane (see ## Upcoming Workstreams).

What We Prove

• The cuneiform benchmark/control artefact chain is real and pinned. Six upstream manifests are recorded by SHA-256 in docs/evidence/ARTEFACT_CHECKSUMS.md and held under HF custody at a pinned revision (docs/HF_CUSTODY_REGISTER.md).
• The inherited governing gate failed and remains failed: P5 closed NO_GO_GOVERNING_GATE_UNMET at governing_1nn_accuracy = 0.021916. P6 (0.038176) and P7 (0.051826) diagnostics are diagnostic-only and never promoted to gate closure.
• SMOKE-01-MANIFEST-VALIDATION runs and passes against the pinned upstream manifest with all 5 cross-field invariants holding and 0 schema errors. See artefacts/smoke/manifest_validation_report.json.
• The smoke surface is stdlib-only, third-party-dependency-free, and reproducible on any Python ≥ 3.8 host without provisioning.

What We Don't Claim

• We do not claim the failed governing gate is repaired. The smoke is custody-only by construction; it cannot move the gate.
• We do not claim cuneiform decipherment, substrate identification, or any positive scientific recovery on this corpus.
• We do not claim public-release readiness, flagship status, or cross-domain superiority.
• We do not claim cleared image rights or rights-cleared corpus redistribution.

Tests and Verification

Surface	Result	Source
Hermetic self-test	pytest -q → 3 passed in 0.04s	tests/test_smoke_runner.py
Real-manifest smoke (pod, scrubbed label)	verdict PASS, observed SHA matches pinned e4d85a…3daa24, 5/5 invariants OK, 0 schema errors	artefacts/smoke/manifest_validation_report.json
HF post-upload SHA verify	6/6 OK against pinned values	artefacts/smoke/hf_upload_verify.json
HF custody (token-verified, latest 2026-04-27)	canonical heavy store on Architect-Prime/cuneiform-control-artefacts (private, rev 3ed3f0d4…, 519 MB total: 6 manifests + orchestrator-added workspace tarballs + tokenizer checkpoints); 6/6 lane SHA + 1 round-trip orchestrator-added SHA verified	docs/HF_CUSTODY_REGISTER.md Verification 5
Local Mac inventory (2026-04-27)	1.4 MB tree; 82 git-tracked + 0 local-only value-bearing files; lane is mac-loss recoverable	docs/HF_CUSTODY_REGISTER.md Verification 5
Operational-leak scan	0 hits for forbidden operational patterns (concrete RunPod IPs, raw SSH endpoints, local home paths, retired monorepo paths, pod IDs)	Operational-leak scan step in .github/workflows/ci.yml
GitHub Actions	minimal private-repo CI on main (this commit)	.github/workflows/ci.yml

Commercial Readiness

Field	Value
Verdict	STAGED
Posture	negative_control_preservation

This lane is public as a lab window, not as a product release. The negative result (NO_GO_GOVERNING_GATE_UNMET) is a first-class scientific output. Commercial readiness in the traditional sense is not the goal of this lane.

Proof Anchors

Need	File
Reviewer landing page	docs/EXECUTOR_STATUS_REPORT.md
Sovereign brief	SOVEREIGN_PRD.md
Failed-gate evidence	docs/evidence/CUNEIFORM_PHASE2_GATE_STATUS.md
Control-results summary	docs/evidence/CUNEIFORM_CONTROL_RESULTS.md
Pinned SHA-256 register	docs/evidence/ARTEFACT_CHECKSUMS.md
Smoke validation report	artefacts/smoke/manifest_validation_report.json
HF upload verify	artefacts/smoke/hf_upload_verify.json
HF custody register	docs/HF_CUSTODY_REGISTER.md
Path-rewrite ledger	docs/PATH_REWRITE_LEDGER.md
Smoke target spec	docs/MINIMAL_SMOKE_TARGET.md
Architecture / truth map	docs/ARCHITECTURE.md
Auditor playbook	AUDITOR_PLAYBOOK.md
Public audit limits	PUBLIC_AUDIT_LIMITS.md
Current state	.gpd/STATE.md

Repo Shape

Path	What Lives Here
NOTICE	Apache-2.0 code license plus CC-BY-4.0 documentation posture.
README.md, SOVEREIGN_PRD.md, AUDITOR_PLAYBOOK.md, AGENTS.md	Front door, sovereign brief, reviewer entry, agent brief.
SOURCE_BOUNDARY.md, DATA_POLICY.md, GOVERNANCE.md, RELEASING.md, PUBLIC_AUDIT_LIMITS.md	Policy surfaces.
code/cuneiform_control/	Stdlib-only smoke runner + JSON Schema.
tests/	Hermetic pytest-discovered self-test (3 cases).
pyproject.toml	Package manifest (Private :: Do Not Upload; zero runtime deps).
artefacts/smoke/	Deterministic smoke and HF-upload-verify reports.
docs/	Architecture, ledgers, smoke spec, HF custody, rerun guide, executor status.
docs/evidence/	Failed-gate evidence, control results, SHA-256 pin register.
docs/migration/	Original migration-package brief, authority, source inventory, handover (preserved as transfer history).
.gpd/	GPD phase pack: project, requirements, roadmap, state, four closed phases (00–03).
.github/workflows/ci.yml	Minimal private-repo CI: install + pytest + smoke help.
_internal/	Agent-orchestration scaffolding (AUTONOMOUS_EXECUTION_POLICY, GPD_BOOTSTRAP_GUIDE, STARTUP_PROMPT, TEMPLATE_USAGE, WORKSTREAM_GPD_INIT_CHECKLIST, MIGRATION_PLAN, TODO). Not reader-facing.

Quick Start

Verified end-to-end on 2026-04-25. Commands run from a fresh clone.

# 1. Install the smoke surface (zero third-party runtime deps; pytest pulled in for tests)
python3 -m venv /tmp/cuneiform-control
source /tmp/cuneiform-control/bin/activate
python -m pip install --upgrade pip
python -m pip install -e . pytest

# 2. Hermetic self-test against bundled fixtures (3 cases, < 1 second)
pytest -q

# 3. Confirm the console entry point works
cuneiform-smoke --help

# 4. Operational-leak scan — runs in CI; reproduce locally by re-running the
#    `Operational-leak scan` step from `.github/workflows/ci.yml`. No concrete
#    IPs, raw SSH endpoints, local home paths, retired monorepo paths, or
#    pod IDs are permitted in tracked files (the workflow itself is excluded
#    because it legitimately defines those patterns).

# 5. Real-manifest replay (only with a locally-held pinned manifest copy):
cuneiform-smoke \
  --manifest /path/to/annotated_sign_benchmark_manifest.json \
  --schema   code/cuneiform_control/schemas/benchmark_manifest.schema.json \
  --checksum e4d85abf3bfa6901a6b20f7c612f1113e77ef9173ca42e00c9867b88b23daa24 \
  --manifest-label "<MONOREPO>/workspace/artifacts/cuneiform/annotated_sign_benchmark_manifest.json" \
  --report /tmp/replay_report.json

A PASS verdict from step 5 proves manifest custody and shape against the pinned SHA-256. It does not repair NO_GO_GOVERNING_GATE_UNMET.

Current Gaps

• Full upstream extraction of cuneiform helper modules (benchmark_common.py, bench_tokenizer.py, probe_1nn.py, P6/P7 diagnostics) remains deferred per docs/PATH_REWRITE_LEDGER.md. The smoke is custody-only, not a full rerun.
• revert_phase2_common.py is UPSTREAM_NOT_PRESENT (S-06); flagged in the checksum register, not a blocker for this lane.
• Image-bearing sources and some upstream datasets remain rights-constrained or fetch-only; pixel-bearing data is never vendored.
• NO_GO_GOVERNING_GATE_UNMET remains the sovereign scientific truth. Repairing it is out of scope for this lane and is external work for a separate workstream (likely gnosis-glyph-engine or gnosis-falsification-harness).

Upcoming Workstreams

Active Engineering

• Smoke surface tightening if new schema variants are contributed (the stdlib-only constraint is a feature, not a bug — preserve it).

Research-Deferred — Investigation Underway

• None currently. If upstream changes substrate-discrimination methodology, revisit whether P6/P7 diagnostics graduate from diagnostic-only to gate candidates.

Operations / External Dependency

• P5 governing-gate repair is owned by a separate workstream (likely gnosis-glyph-engine or gnosis-falsification-harness). This lane preserves the negative result; it does not initiate gate repair.
• Rights clearance for image-bearing corpus redistribution remains unresolved; no action here until upstream rights review completes.

Zero-Base Scientific Thinking — GPD Research and Planning Pending

• None currently scheduled. A new milestone in this lane opens only if a separate workstream produces gate-repair evidence that warrants revisiting the governing verdict.