Architecture | Readiness | API | Benchmarks | Legal | Package

SAL v6.2 — free below $100M annual revenue. See LICENSE.

Decode-deterministic sensor codec with 9 domain presets. 6.6× bounded-lossy compression. 27/27 destructive tests passed. Edge-deployable. Sensor deltas are encoded using an 8-direction amplitude gear codebook with log-magnitude quantisation and RLE.

ZPE-IoT is a deterministic sensor compression SDK for constrained IoT streams — built for industrial IoT platform teams and edge telemetry vendors where transmission bandwidth is expensive, storage budgets are fixed, and lossy black-box codecs are unacceptable. Rust core, Python bindings via PyO3. Every metric traces to committed artifacts under validation/ and proofs/.

The repo is private-stage. Install path and proof artifacts are real. Published on PyPI as zpe-iot. Local arm64 macOS wheel install is verified; multi-platform wheel closure is not claimed here.

Not claimed: universal compressor dominance, lossless reconstruction, runtime coupling to ZPE-IMC, or protocol/runtime closure beyond the published package surface.

Source: validation/results/bench_summary_E1_real_public_20260321T225305.json, validation/results/release_preflight_report_20260321T205127.json, validation/results/dt_results_20260321T225304.json

Competitive benchmark evidence: proofs/artifacts/public_benchmarks/INDEX.json

Framing disclosure: ZPE-IoT is a bounded-lossy codec (NRMSE ≤ 0.004, bit_parity=false). All comparators below (zstd, zlib, LZ4, Gorilla-proxy) are lossless. Direct ratio comparison is informative but not apples-to-apples — ZPE-IoT achieves higher compression in part because it tolerates bounded reconstruction error that lossless codecs do not permit. Margins below ~1.5× (e.g. DS-05 NOAA: ZPE 7.29× vs zlib 7.02×) may fall within the lossy advantage alone.

Comparators above use default or moderate compression levels; at max compression (brotli-11, zstd-22), lossless baselines can exceed ZPE-IoT on individual datasets.

DS-12 outlier disclosure: DS-12 is a high-redundancy dataset where general-purpose compressors vastly outperform ZPE-IoT (e.g. zstd achieves 5957.82× vs ZPE's 120.47×). Including DS-12 inflates ZPE's mean to 17.16× but inflates competitors even more, making the all-datasets mean misleading for both sides. The headline 6.65× (DS-01..DS-10) is the honest comparison surface. ZPE-IoT does not claim universal compressor dominance.

Gorilla-proxy disclosure: The "Gorilla-proxy" comparator is a simplified ~25-line XOR-delta + zlib implementation inspired by Facebook Gorilla's XOR encoding approach. It is not Facebook's production Gorilla time-series codec. See validation/benchmarks/bench_vs_gorilla.py for the full implementation.

Baseline methodology: All compression ratios use float64 (8 bytes/sample) as the raw-size denominator. Against a float32 (4 bytes/sample) baseline, ratios would be approximately half the reported values. ZPE-IoT is a bounded-lossy codec; general-purpose baselines (zstd, LZ4, zlib) are lossless.

Auditable guarantees backed by committed proof artifacts. Start at validation/results/IOT_WAVE1_RELEASE_READINESS_REPORT.md.

• 6.65× mean compression across DS-01..DS-10 (10 non-outlier sensor datasets); DS-12 excluded from headline — see Competitive Benchmarks
• 27/27 destructive tests passed
• Decode-deterministic replay on tested corpus
• Managed preflight 17 PASS / 0 FAIL / 1 DEFERRED
• Fresh install smoke test PASS on arm64 macOS

• No claim of lossless reconstruction (bounded-lossy codec)
• Broad deployment/runtime closure beyond the published package surface
• No claim of EnOcean or proprietary protocol support
• No claim of MQTT/LoRaWAN production bridge
• No claim of direct Gorilla parity — our Gorilla-proxy comparator is a simplified XOR+zlib implementation, not Facebook's production Gorilla codec
• No claim of unlimited stream length — codec has a 65,536-sample hard cap (2-byte header); approximately 16 minutes at 60 Hz
• No claim of NaN/Inf tolerance — non-finite floating-point inputs cause codec failure
• No claim of signal-level roundtrip idempotency — as a bounded-lossy codec, encode(decode(encode(x))) ≠ encode(x) is expected; determinism applies to the decode path (same packet → same output)

Evaluators: pip install zpe-iot closes the public package surface. For repo-local evaluation, use python -m pip install -e './python[dev]' in a clean venv. Contact [email protected] for integration guidance.

Confidence is derived from the managed-preflight completeness score in validation/results/release_preflight_report_20260321T205127.json: 17 / 18 = 94.4%.

Managed preflight is the build/install/release gate, strict DT is the destructive-test gate, and E1 is the promoted real-public benchmark tier.

Modality Lanes counts the nine preset lanes exposed by python/zpe_iot/presets.py.

Install from PyPI (recommended):

pip install zpe-iot

Install from source:

git clone https://github.com/Zer0pa/ZPE-IoT zpe-iot
cd zpe-iot
python -m pip install -e './python[dev]'
cargo test --manifest-path core/Cargo.toml --release
python validation/destruct_tests/run_all_dts.py --strict-gates

Treat project_docs/ and older release/RC_* bundles as lineage. Current repo truth lives in the cited March 21 proof and validation artifacts above.

This package is part of the Zer0pa ZPE codec portfolio. See also: zpe-xr, zpe-robotics, zpe-geo, zpe-finance, zpe-ink, zpe-multimodal, zpe-neuro, zpe-mocap, zpe-prosody, zpe-bio.

Observability: Comet dashboard (public)