ZPE-Ink

ZPE-Ink is an always-in-beta .zpink digital-ink codec — one of 17 independent encoding products in the Zer0pa portfolio. Every claim in this README is anchored to a committed proof artifact or CI test in this repository.

License: see LICENSE.

What This Is

ZPE-Ink encodes and decodes deterministic .zpink stroke packets in Python, with 0.0 Hausdorff error on all measured datasets and bit-exact roundtrip fidelity backed by committed benchmark artifacts. The public proof surface covers:

• lossless encode/decode roundtrip for generated stroke fixtures
• CRC and truncated-payload rejection
• optional pressure, tilt, and azimuth channel handling
• static binding-contract consistency across the Python, PyO3, WASM, Swift, and C# surfaces

The committed public benchmark artifacts are in proofs/ and their compression-ratio results are surfaced below. CI does not rerun those external corpora; the benchmark rows are static committed artifacts.

Encoding Contract

Public Benchmark Results

These rows are committed static artifacts. The codec ran encode → decode → verify on each dataset using the repo-local lossless path. CI does not rerun these external corpora; results are bounded to the sample sizes shown.

Baseline: raw little-endian float32 x/y pairs per point. Hausdorff = 0.0 on all measured datasets means decoded coordinates match the source integers exactly.

Zero-channel suppression improvement (committed): auto-suppressing zero tilt/azimuth streams raised CROHME mean compression from 1.52× to 1.76× (max 3.34×) and MathWriting mean from 1.06× to 1.15×. This change is captured in proofs/artifacts/mathwriting_analysis/comparison.json and exercised by code/tests/test_codec_roundtrip.py::test_zero_optional_channels_are_omitted_by_default.

Encode latency on the measured corpora: median 0.02–0.10 ms/stroke (single-core Python, macOS; QuickDraw low end 0.026 ms/stroke, MathWriting high end 0.099 ms/stroke). Source: proofs/artifacts/public_benchmarks/dataset_matrix.json (median_ms_per_stroke field).

These results do not constitute a hard-corpus pass, release-readiness claim, or competitive superiority claim. See proofs/artifacts/public_benchmarks/README.md for full methodology notes.

Commercial Readiness

What We Do Not Claim

• No claim of release readiness
• No claim of blind-clone closure
• No claim of hard-corpus pass
• No claim of general digital-ink dominance
• No claim that the public benchmark rows close release readiness or hard-corpus authority
• No claim that local binding-contract checks prove full runtime parity for every downstream environment
• No claim that committed compression ratios on the above datasets constitute superiority over general-purpose codecs on those corpora

Quick Start

Development install:

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

Package build:

python -m build

Repository Links