ZPE-Ink

Deterministic digital-ink codec centered on the .zpink packet format. This repo is a private staging snapshot with a current proof subset and rerun surface. It is not release-ready.

What This Is

Cross-runtime deterministic ink codec. 1.0–2.8× compression on bounded real public corpora, with an additional UJI Pen Characters row at 1.6111× exact roundtrip; 5.59× on structured tier vs raw float32 xy-only baseline. The .zpink packet format encodes stroke streams — pressure, tilt, timing — identically across every runtime. Same input, same output, iOS to Web to Android. Strokes are tokenised using 8-direction Freeman chain codes packed as 4-bit nibbles.

ZPE-Ink targets annotation-runtime teams and cross-platform pen-input infrastructure where ink fidelity matters and generic codecs destroy structural detail. Validated through Calliar non-Latin corpus. Bindings surface: Rust, WASM, Swift, C#.

Readiness: private-stage. Structured-tier transport proved. Sovereign release surface remains FAIL.

Field	Value
Architecture	STROKE_MANIFOLD
Encoding	INK_DELTA_V1

Key Metrics

Metric	Value	Baseline
STRUCT_CR	5.59×	structured tier, synthetic data, vs raw float32 xy-only baseline
REAL_CR	1.02–2.77×	5 public datasets (MathWriting, CROHME, QuickDraw, DigiLeTs, Calliar)
UJI_PUBLIC	1.6111×	exact roundtrip on bounded UJI Pen Characters public row
PARITY	3/3	Py/Rust/WASM decode

Competitive Benchmarks

Structured tier; version-locked tools. Source: proofs/reruns/benchmark_freeze_local/

Tool	Ratio	Notes
Brotli (q11)	6.83×	wins structured tier
ZPE-Ink	5.59×	5 runtimes; deterministic
zstd (l19)	4.92×	—
LZ4 (l9)	1.99×	—

Brotli wins on ratio; ZPE-Ink wins on cross-runtime parity and deterministic encode.

What We Prove

Auditable guarantees backed by committed proof artifacts. Start at proofs/reruns/phase5_wedge/final_go_no_go_surface.json.

• Structured-tier compression exceeds 5× vs raw float32
• Python/Rust/WASM decode parity passes locally
• Swift/C# header contracts pass (magic, version, header size)
• Structured tier ratio: 5.59×
• UJI Pen Characters public row: 1.6111× with exact roundtrip on 1,364 samples / 74,592 points

What We Don't Claim

• No claim of release readiness (release surface FAIL)
• No claim of blind-clone closure (INCONCLUSIVE)
• No claim of hard-corpus pass
• No claim of general digital-ink dominance
• 5× compression on real handwriting data — structured-tier 5.59× applies to synthetic data against a raw float32 xy-only baseline; real public corpora achieve 1.0–2.8×
• No claim that the UJI Pen Characters row closes release readiness, hard-corpus authority, or the sovereign gate

Commercial Readiness

Field	Value
Verdict	INCONCLUSIVE
Release posture	Live work in progress; not a final official release
Commit SHA	98b5ed734735
Confidence	67%
Source	proofs/reruns/phase5_wedge/final_go_no_go_surface.json

Evaluators: Proof surface available for inspection. Sovereign surface FAIL — see Open Risks. Contact hello@zer0pa.com.

Current authority reflects the latest committed verification surface. It does not imply release readiness. Confidence basis: 4/6 verification checks currently PASS.

Tests and Verification

Code	Check	Verdict
V_01	Structured-tier compression boundary	PASS
V_02	Python/Rust/WASM parity	PASS
V_03	Swift/C# header contracts	PASS
V_04	Pytest regression surface	PASS
V_05	Sovereign release surface	FAIL
V_06	Blind-clone closure	INC

Proof Anchors

Path	State
proofs/reruns/phase5_wedge/final_go_no_go_surface.json	CURRENT
proofs/reruns/benchmark_freeze_local/claim_scope_map.json	CURRENT
proofs/reruns/contradiction_resolution_local/contradiction_resolution_manifest.json	CURRENT
proofs/reruns/phase3_public_benchmarks/phase3_public_benchmarks.json	CURRENT
proofs/logs/20260321_technical_alignment_cross_runtime.json	VERIFIED
proofs/logs/20260321_technical_alignment_binding_contracts.json	VERIFIED
proofs/logs/20260321_technical_alignment_pytest.txt	VERIFIED
proofs/logs/20260321_technical_alignment_wheel_install.txt	VERIFIED

Any contradiction across these anchors keeps the repo INCONCLUSIVE.

Repo Shape

Field	Value
Proof Anchors	8
Modality Lanes	6
Authority Source	proofs/reruns/phase5_wedge/final_go_no_go_surface.json

Quick Start

# Install from PyPI
pip install zpe-ink

Or install from source (development). Prereqs: Python 3.11+, Rust toolchain, and wasm32-unknown-unknown target for binding checks.

python -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -e './code[dev]'
python -m pytest code/tests -q
python -m zpe_ink demo
python -m zpe_ink verify-roundtrip

Field	Value
Repository URL	https://github.com/Zer0pa/ZPE-Ink
Contact	architects@zer0pa.ai

Verification anchors: proofs/logs/20260321_technical_alignment_pytest.txt, proofs/logs/20260321_technical_alignment_wheel_install.txt.

Ecosystem

• ZPE-IMC - reference repo for shared documentation layout and proof-surface conventions.
• ZPE-Mocap - adjacent motion-stream codec in the ZPE transport family.
• ZPE-XR - sibling XR motion compression surface with multi-runtime packaging work.

Observability: Comet dashboard (public)

Route	Target
Documentation index	docs/ARCHITECTURE.md
Auditor path	proofs/reruns/phase5_wedge/final_go_no_go_surface.json
Governance rules	GOVERNANCE.md
Release gate rules	RELEASING.md
Contribution workflow	CONTRIBUTING.md
Security policy	SECURITY.md
Support routing	docs/LEGAL_BOUNDARIES.md

Who This Is For

Ideal first buyer	Stylus/annotation runtime team or cross-platform pen-input infrastructure team
Pain	Ink streams vary across runtimes — iOS, Web, Android — requiring per-platform codecs with no determinism guarantee
Deployment	SDK — Python package with repo-local Rust/WASM/Swift/C# bindings
Family position	Secondary product candidate in the Zer0pa deterministic encoding family. Not the lead commercial front door