zpe-neuro 0.1.1

Deterministic spike-event extraction codec for electrophysiology signals; Python; Gate C/D PASS on bounded DANDI 000034

ZPE-Neuro

What This Is

ZPE-Neuro is the extracellular recording lane of the Zer0pa 17-lane encoding portfolio — a bounded spike-event extraction codec for electrophysiology signals. It is independent of other portfolio lanes and speaks only for its own domain.

The strongest CI-anchored result to date: deterministic encode-decode round-trip on DANDI 000034 with a 401x event ratio, 78.44 µV RMSE, and Gate C + Gate D both PASS on blind-clone replay from the current origin/main snapshot. No comparison baseline exists for this lane; the numbers stand on their own terms.

This front door promotes only claims backed by a tracked proof artifact and exercised in CI. Treat CURRENT_AUTHORITY_PACKET.md as the April 24 routing layer; the full proof archive goes deeper.

Current Verified Surface

Claim	Proof artifact	CI coverage
DANDI 000034 remains the positive public anchor with 41 events, 401.04x event ratio, and 78.44 uV RMSE.	public_corpus_eval_dandi_000034_mouse412804_ecephys.json	tests/test_dandi_offline.py
Blind-clone replay from current repo truth closed with Gate C and Gate D both PASS.	verification_summary.md, gate_c_summary.json, gate_d_summary.json	Verify Package Surface / proof-import-smoke, gate-slice, tests/test_roundtrip.py, tests/test_wave1_determinism.py
Breadth adjudication records IBL as the counted second extracellular target and does not count the Tier 1 DANDI anchor as breadth closure.	public_corpus_summary.json, public_corpus_ibl_waveform_eval.json	tests/test_breadth_adjudication.py
DANDI 000003 was executed as the first next-family DANDI breadth probe and recorded FAIL.	public_corpus_eval_dandi_000003_yutamouse20_ecephys.json, dandi000003_decision.md	tests/test_breadth_adjudication.py
AJILE12 remains explicitly out of family and is excluded from counted breadth.	ajile12_family_boundary_decision.md, public_corpus_summary.json	tests/test_breadth_adjudication.py

Current Metrics

DANDI 000034 Tier-1 Authority Anchor

Metric	Value	Proof artifact	CI test
Event ratio (window-scoped)	401x	benchmark_summary.json	tests/test_dandi_offline.py::test_fixture_reproduces_benchmark_metrics
RMSE	78.44 uV	benchmark_summary.json	tests/test_dandi_offline.py::test_fixture_reproduces_benchmark_metrics
Encode latency (mean / max)	0.089 ms / 0.208 ms	benchmark_summary.json	artifact only — no pytest bound asserted
Decode latency (mean / max)	0.474 ms / 0.686 ms	benchmark_summary.json	artifact only — no pytest bound asserted

These are window-scoped metrics (6000-sample, 8-channel window at 30 kHz). They are not whole-recording compression results.

IBL Second-Target (Tier-2 Breadth, Counted PASS)

Metric	Value	Proof artifact	CI test
Event ratio (window-scoped)	224x	public_corpus_ibl_waveform_eval.json	artifact only — tests/test_breadth_adjudication.py tests logic, not this metric value
RMSE	38.16 uV	public_corpus_ibl_waveform_eval.json	artifact only — tests/test_breadth_adjudication.py tests logic, not this metric value

Gate D: Embedded Latency and Drift Resilience

Metric	Value	Proof artifact	CI test
Modeled latency (mean / p99)	612.5 ns / 850 ns	neuro_embedded_latency.json	CI gate-slice
Latency threshold	< 900 ns	neuro_embedded_latency.json	CI gate-slice
Drift accuracy at 0–15 µm	100%	neuro_drift_resilience.json	CI gate-slice
Drift cliff	at 20 µm	neuro_drift_resilience.json	CI gate-slice

Embedded latency uses a hardware-proxy cycle model at 80 MHz ARM-class clock plus Python reference timing. It is not a measured on-silicon result.

Determinism

Metric	Value	Proof artifact	CI test
Identical-hash runs	5 / 5 seeds	determinism_replay_results.json	tests/test_wave1_determinism.py, tests/test_roundtrip.py
NWB roundtrip SHA256	bit-consistent	neuro_nwb_roundtrip.json	tests/test_roundtrip.py

What We Don't Claim

• No claim of lossless signal reconstruction.
• No claim that the window-scoped event-encoding ratios are whole-recording compression results.
• No claim that DANDI 000003 closed new breadth.
• No claim of commercialization-safe closure or tagged public release.
• No claim beyond the bounded extracellular lane.

Repo Shape

Field	Value
Proof Anchors	5
Modality Lanes	1
Authority Source	proofs/manifests/CURRENT_AUTHORITY_PACKET.md

• src/zpe_neuro/: installable extractor package.
• tests/: repo-local verification slice.
• tools/: gate runners and operator scripts.
• proofs/: current authority packet plus April 24 replay and breadth packets.
• docs/: architecture, legal boundaries, release status, and dataset-scope notes.

Quick Start

git clone https://github.com/Zer0pa/ZPE-Neuro.git
cd ZPE-Neuro
python3.11 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -e '.[dev]'
python -m pytest tests

For the bounded gate slice:

python -m pip install -e '.[gate,proof]'
python tools/run_gate_c.py --artifact-root artifacts/manual_gate_c --seed 20260220
python tools/run_gate_d.py --artifact-root artifacts/manual_gate_d --replay-seeds 20260220,20260221,20260222,20260223,20260224

Read docs/LEGAL_BOUNDARIES.md before widening any claim from this repo state.