Compress movement traces and keep them spatially searchable. Fleet routes, vessel tracks, AV telemetry, logistics trajectories — indexed during encoding so downstream lookup does not need the raw stream.

ZPE-Geo is deterministic trajectory compression with H3 hexagonal spatial indexing, a maneuver-search surface on the compressed representation, and fidelity validation. Trajectories are encoded as sequences of 8-compass direction tokens with magnitude quantisation and RLE compression. The strongest current wedge is storage plus spatial indexing: the committed AV maneuver benchmark now measures p@10 mean 0.33 (left_turn 0.0, lane_merge 0.0, stop 1.0), so this repo should not currently be read as a strong maneuver-ranking result.

Zer0pa SAL is free below $100M annual revenue. See LICENSE.

Sources: AIS_CR, AV_CR, and ENCODE_P95 remain historical-only Wave-1 metrics from proofs/artifacts/2026-02-20_zpe_geo_wave1/.

SEARCH was recomputed on 2026-04-15 on the committed benchmark path in code/scripts/gate_c_benchmarks.py against code/fixtures/av_argoverse2_fixture_v1.json after removing benchmark label injection from ManeuverSearchIndex.build().

The March 21 operator pack remains the governing release-readiness surface.

Wave-1 AIS corpus (190 trajectories). Source: proofs/artifacts/2026-02-20_zpe_geo_wave1/

Framing disclosure: ZPE-Geo is a lossy trajectory codec (lossy at quant_step=0.25m; max coordinate error 0.0018° ≈ 200m at equator on real NOAA AIS data). Douglas-Peucker is also lossy, so the comparison below is lossy-vs-lossy. Neither codec preserves lossless coordinates. On real-world NOAA AIS extracts, coordinate_exact_match_count = 0/5. See proofs/artifacts/real_world_benchmarks/noaa_ais_day_extract_benchmark.json.

ACM 2025: doi:10.1145/3764920.3770598. Direct dataset parity with paper corpus INCONCLUSIVE.

Auditable guarantees backed by committed proof artifacts. Start at proofs/artifacts/2026-03-21_operator_status/README.md.

• Trajectory compression with preserved spatial query capability
• H3 hexagonal spatial indexing during encoding
• Maneuver search surface on compressed representation; current committed AV benchmark is p@10 mean 0.33
• Lightweight test suite passes

• No claim of blind-clone closure (GEO-C001)
• No claim of full-corpus closure (GEO-C002)
• No claim of release readiness (GEO-C004)
• No claim of strong maneuver-retrieval quality — the committed AV benchmark path currently measures p@10 mean 0.33, with left_turn and lane_merge at 0.0
• No claim of superiority over incumbent geospatial compression
• Lossless coordinate preservation — compression at default settings (quant_step=0.25m) introduces up to 0.0018° (~200m at equator) coordinate error. On real NOAA AIS data, coordinate_exact_match_count = 0/5. See proofs/artifacts/real_world_benchmarks/noaa_ais_day_extract_benchmark.json
• Real-corpus equivalence for simulated query benchmarks — the 10M-corpus query-latency figure in historical proofs uses replicated synthetic trajectories, not a real-world corpus

Evaluators: Proof surface available for inspection. See Open Risks for remaining gaps. Contact [email protected].

• Supporting operator pack: proofs/artifacts/2026-03-21_operator_status/README.md
• Open gates: GEO-C001, GEO-C002, GEO-C004
• Confidence basis: 5 / 8 tracked claims green on proofs/artifacts/2026-03-21_operator_status/phase0311_runpod/max_claim_resource_map.json

Quickest outsider orientation:

pip install zpe-geo

The steps below verify the current repo-local package surface. They do not prove blind-clone closure, full-corpus closure, or release readiness.

git clone https://github.com/Zer0pa/ZPE-Geo.git zpe-geo
cd zpe-geo
python -m venv .venv
source .venv/bin/activate
python -m pip install -e ".[dev,h3]"
python -m pytest code/tests -q
python - <<'PY'
from zpe_geo import H3Bridge, ManeuverSearchIndex, decode_trajectory, encode_trajectory
print("zpe-geo import OK")
print("h3 backend:", H3Bridge().backend)
print("search surface:", ManeuverSearchIndex.__name__)
PY

After a successful repo-local verification you should have:

• an editable install of the repo-root package surface
• passing lightweight repo-local tests under code/tests
• an importable zpe_geo surface without relying on outer-workspace material

• Repo-root package metadata uses LicenseRef-Zer0pa-SAL-6.2.
• LICENSE remains the in-repo legal text shipped with this checkout.
• Commercial or hosted use above the SAL threshold requires contact at [email protected].
• LICENSE is the legal source of truth. Repo docs summarize it; they do not override it.

Observability: Comet dashboard (public)

The archived bundle under proofs/artifacts/2026-02-20_zpe_geo_wave1/ remains part of the repo because it contains real historical evidence:

• archived performance metrics across all eight promoted claims
• archived comparator notes, including an in-repo AIS baseline comparison
• preserved contradictions that explain why archived success does not equal current release authorization

Read those facts as historical-only context through the archived wave-1 bundle, not as current release status.