ZPE-Mocap is a deterministic motion fingerprinting and retrieval codec for skeletal motion data. The commercial wedge is not faithful pose reconstruction. It is compact motion fingerprints for search, indexing, and retrieval where raw BVH archives are expensive to store and slow to query.

The real-data authority surface is the committed CMU fixture benchmark: 18.77× mean bandwidth reduction, 32.45 mm mean MPJPE, and 82.51° mean joint-angle RMSE across 10 BVH clips. That angle error is too high for playback-grade reconstruction, so the lane must be read as retrieval/indexing only. The older synthetic bundle remains useful for ceiling behavior and search latency, but it is not the commercial front door.

Source: results.json, summary.md, mocap_search_eval.json

The only committed comparator tables are synthetic. They are useful for lineage, but they are not the commercial wedge because the synthetic corpus is generated from the ZPE token vocabulary itself. Real evaluation should be judged from the CMU fixture results above.

Source: acl_direct_comparator_table.json, mocap_compression_benchmark.json

• 18.77× mean bandwidth reduction on the committed 10-clip CMU fixture corpus.
• 32.45 mm MPJPE and 82.51° joint-angle RMSE on real BVH data, which is sufficient for retrieval/indexing judgment but not playback.
• Synthetic search ranking at p@10 = 1.0 and synthetic query latency at 26.14 ms p95.
• A committed Wave-1 evidence bundle with explicit commercialization and integration boundaries.

• No claim of playback-quality pose reconstruction.
• No claim of real-data parity with the synthetic benchmark bundle.
• No claim that the synthetic ACL comparison is a fair commercial benchmark.
• No claim of Blender runtime closure, clean-clone verification, or commercialization-safe closure.
• No claim that this repo is a released animation-runtime product.

The live commercial wedge is motion fingerprinting and similarity retrieval. Anything broader than that is overstating what the current evidence supports.

• code/: Python reference implementation and repo-local tests.
• proofs/artifacts/2026-02-20_zpe_mocap_wave1/: historical synthetic ceiling bundle.
• proofs/artifacts/2026-04-14_cmu_corpus_benchmark/: current real-data benchmark authority.
• docs/: architecture and legal-boundary notes.
• proofs/logs/, proofs/source_refs/: lineage and source-reference material.

# Install from PyPI
pip install zpe-mocap

Or use the repository verification path:

git clone https://github.com/Zer0pa/ZPE-Mocap.git
cd ZPE-Mocap
python -m venv .venv
source .venv/bin/activate
python -m pip install -e ./code
python -m unittest discover -s code/tests -v

Smoke check:

python - <<'PY'
from zpe_mocap.codec import decode_zpmoc, encode_clip
from zpe_mocap.synthetic import generate_clip

clip = generate_clip(
    clip_id="readme_smoke",
    label="walk",
    frames=120,
    fps=60,
    seed=20260220,
    noise_scale=0.0002,
)
enc = encode_clip(clip, seed=20260220)
dec = decode_zpmoc(enc.payload)
print(enc.compression_ratio, dec.clip_id)
PY

Read docs/LEGAL_BOUNDARIES.md before turning any synthetic result into a broader commercial claim.