zpe-ft 0.1.0

ZPE financial time-series codec and pattern search

19.19x controlled-corpus OHLCV compression. 20.57x controlled-corpus tick compression. Repo install verified.

python -m pip install -e .

build log

clean install log

pytest log

---

At A Glance

Deterministic market-data codec. OHLCV bars. Tick streams. Pattern search.

30 real-market equity series benchmarked. 3 real-market tick proxy series benchmarked. Optional Rust helper.

Wave-1 proof retained. Install surface verified. Benchmark boundary tracked.

Quant researcher. Tick-history archive. Market replay engineer. Backtest pipeline owner.

vs Parquet: column store, no native search. vs kdb+: proprietary stack. ZPE-FT: deterministic compression with native pattern search.

Anchor	Artifact
Install evidence	20260321T202948Z_clean_install_verify.log
Controlled proof report	CONSOLIDATED_PROOF_REPORT.md
Real-market boundary	BOUNDARY.json

Quick Start

python -m venv .venv
source .venv/bin/activate
python -m pip install -e .
python -c "import zpe_finance; print(zpe_finance.__all__[:4])"

---

Quick Start

Quick Verify

Repo checkout verification. Current import surface. Current evidence logs.

git clone https://github.com/Zer0pa/ZPE-FT.git
cd ZPE-FT
python -m venv .venv
source .venv/bin/activate
python -m pip install -U pip
python -m pip install -e .
python - <<'PY'
import zpe_finance
from zpe_finance.rust_bridge import rust_version
print("exports", sorted(zpe_finance.__all__))
print("rust_bridge", rust_version())
PY

Expected outputs:

• the package imports cleanly from the repo root
• the exported surface comes from python/zpe_finance/
• rust_bridge truthfully reports python-fallback unless you explicitly build the optional helper under core/
• shipped install and pytest evidence remains at proofs/artifacts/operations/20260321T202948Z_build.log, proofs/artifacts/operations/20260321T202948Z_clean_install_verify.log, and proofs/artifacts/operations/20260321T202948Z_pytest_alignment.log

Fastest evidence path:

docs/AUDITOR_PLAYBOOK.md

docs/PUBLIC_AUDIT_LIMITS.md

docs/ARCHITECTURE.md

License Boundary

• Free to use at or below USD 100M annual gross revenue under SAL v6.2.
• SPDX tag: LicenseRef-Zer0pa-SAL-6.0.
• Commercial or hosted use above that threshold requires contact at [e-mail].
• LICENSE is the legal source of truth; CHANGELOG.md and CITATION.cff are release metadata, not substitute license terms.

What This Is

Store and replay financial time-series with deterministic fidelity. Search compressed OHLCV streams without decoding them.

ZPE-FT is a codec and pattern-search SDK targeting market-data infrastructure teams and quant research platforms. Installable Python package, optional Rust helper, retained Wave-1 proof bundles on controlled corpus, and validated delayed-feed acquisition on SPY/AAPL.

Persona	Why this repo matters
Quant researcher archiving tick history	Compress delayed-feed history, retain deterministic decode, and search motifs without handing storage over to a proprietary incumbent.
Market replay engineer	Package bars and top-of-book streams into a replayable format with bounded fidelity and query hooks for backtest and simulation pipelines.

Question	Answer
What is this?	A source-available repo for a financial time-series codec, query surface, and proof packet. It is real software with retained evidence, not a marketing shell.
What is the current authority state?	Three surfaces matter: shipped package/build/install evidence, the retained controlled Wave-1 codec bundle, and the current Phase 06 blocker packet for the open-access enterprise benchmark.
What is actually proved?	The repo-native package path works, the carried Wave-1 bundle supports bounded codec/search claims on the controlled corpus, and the delayed-feed acquisition/freeze/refresh path is real on a bounded SPY/AAPL smoke.
What is not being claimed?	No claim of public-release readiness, open-access enterprise benchmark closure, Timescale equivalence, or broad incumbent displacement.
Where should an outsider start?	Clone from https://github.com/Zer0pa/ZPE-FT.git, run the quick verify steps above, then read docs/AUDITOR_PLAYBOOK.md, docs/PUBLIC_AUDIT_LIMITS.md, and proofs/FINAL_STATUS.md together.

Field	Value
Architecture	MARKET_STREAM
Encoding	OHLCV_DELTA

Key Metrics

Metric	Value	Source
OHLCV_CR (Wave-1 controlled)	19.19×	ft_ohlcv_benchmark.json
TICK_CR (Wave-1 controlled)	20.57×	ft_tick_benchmark.json
OHLCV_CR (provider-max 1m proxy)	14.16×	Non-authority proxy; see MISSING_DATA_README.md
TICK_CR (Dukascopy tick proxy)	11.11×	Non-authority proxy; see MISSING_DATA_README.md
VS_PARQUET (Binance public)	2.81×	PHASE3_PUBLIC_BENCHMARKS.json
DATASETS	33	30 equity + 3 tick series

Authority source: Wave-1 controlled corpus at proofs/artifacts/2026-02-21_ft_wave1_final/. Phase 3 public benchmarks at BENCHMARKS.md. Proxy lane results are non-authority rehearsal artifacts.

Competitive Benchmarks

Wave-1 controlled comparators: proofs/artifacts/2026-02-21_ft_wave1_final/. Parquet comparison: PHASE3_PUBLIC_BENCHMARKS.json.

2.81× vs Parquet+ZSTD on Binance BTCUSDT aggTrades (Phase 3 public benchmark); search-on-compressed, deterministic replay.

Tool	Ratio (raw)	Search
bz2	16.12×	—
lzma	13.13×	—
zlib/gzip	8.35×	—

zstd/lz4 unavailable at measure time. No general-purpose tool offers search-on-compressed or deterministic replay.

What We Prove

Auditable guarantees backed by committed proof artifacts. Start at AUDITOR_PLAYBOOK.md.

• Deterministic financial time-series encoding (OHLCV + ticks)
• Pattern search and market replay on compressed format
• Repo-local install verifies and pytest surface passes
• Comet observability hooks wired for proof tracking

What We Don't Claim

• No claim of public release readiness
• No claim of Phase 06 contract closure (blocked on missing inputs)
• No claim of real-time trading system integration
• No claim of regulatory compliance (market data use)
• Provider-max proxy lane results (14.16x OHLCV, 11.11x tick) are non-authority rehearsal artifacts and cannot be promoted into Phase 06 evidence
• Real-market boundary (BOUNDARY.json) shows phase06_contract_match: false and minimum OHLCV compression below the 10x authority floor on some proxy lanes

Commercial Readiness

Field	Value
Verdict	BLOCKED_MISSING_INPUTS
Commit SHA	7394b4bd04e0
Confidence	67%
Source	proofs/FINAL_STATUS.md

Evaluators: Phase 06 contract freeze blocked on missing inputs. Evaluation surface available — consult Open Risks before integration planning. Contact [e-mail].

Tests and Verification

Code	Check	Verdict
V_01	Package build / install / import truth	PASS
V_02	FT-C001 OHLCV compression	PASS
V_03	FT-C002 tick compression	PASS
V_04	FT-C004 pattern search	PASS
V_05	Phase 06 contract freeze	FAIL
V_06	DB breadth beyond SQLite	INC

Authority Notes

Package / install truth SUPPORTED Current build, clean install, import, and pytest alignment logs are shipped under proofs/artifacts/operations/.

Controlled codec authority 2026-02-21_ft_wave1_final Promoted codec/search claims remain tied to the carried Wave-1 controlled corpus bundle.

Benchmark gate posture BLOCKED_MISSING_INPUTS The open-access enterprise benchmark still lacks `33` named corpus exports plus the auditable query catalog and truth labels.

Surface	Locked value	Why it matters
Repository URL	https://github.com/Zer0pa/ZPE-FT.git	This is the live public acquisition surface for the workstream.
Current controlled bundle	proofs/artifacts/2026-02-21_ft_wave1_final/	The promoted OHLCV, tick, fidelity, query-latency, and SQLite roundtrip claims stay bounded to this carried bundle.
Current blocker packet	proofs/reruns/2026-03-21_phase06_contract_freeze_attempt_v3/missing_inputs_packet.json	This is the sovereign open-access enterprise benchmark truth until the missing inputs exist and a clean freeze/replay closes.
Bounded market smoke	proofs/reruns/2026-03-19_alpaca_demo_smoke/	Shows the repo-native delayed-feed acquisition, freeze, and refresh path is real on SPY and AAPL, but does not close the enterprise gate.
Current release posture	NOT_PUBLIC_READY	The repo can be cloned and verified, but the release gate remains open.
DB posture	SQLite supported; Timescale INCONCLUSIVE	Current DB truth is narrower than general incumbent-displacement claims.

Three Dimensions Of Authority

Package Surface The install/import/build path is real and auditable from the repo root.

Controlled Codec Evidence Performance and pattern-search claims are bounded to the retained Wave-1 controlled corpus bundle.

Benchmark Honesty The repo carries a named blocker packet instead of narrating an unearned pass on the open-access enterprise benchmark.

Proof Anchors

All promoted FT values should be read through the package/build/install logs, the retained controlled Wave-1 bundle, and the blocker packet together.

Path	State
proofs/FINAL_STATUS.md	VERIFIED
proofs/CONSOLIDATED_PROOF_REPORT.md	VERIFIED
proofs/artifacts/operations/20260321T202948Z_clean_install_verify.log	VERIFIED
proofs/artifacts/2026-02-21_ft_wave1_final/	VERIFIED
proofs/reruns/2026-03-21_phase06_contract_freeze_attempt_v3/missing_inputs_packet.json	VERIFIED

Anchor Details

proofs/artifacts/operations/20260321T202948Z_build.log Build and package preparation proof for the current repo surface.	proofs/artifacts/operations/20260321T202948Z_clean_install_verify.log Clean install and import proof for the public clone path.
proofs/artifacts/operations/20260321T202948Z_pytest_alignment.log Pytest alignment log for the shipped Python package surface.	proofs/CONSOLIDATED_PROOF_REPORT.md Claim-by-claim reading guide covering bounded codec evidence and unresolved surfaces.

Proof rung	Locked value	What it proves now
Package / install proof	SUPPORTED	The public clone path installs and imports from the repo surface as documented.
Controlled codec bundle	2026-02-21_ft_wave1_final	Carries the promoted OHLCV, tick, fidelity, pattern-search, latency, and SQLite roundtrip claims on the controlled corpus.
Delayed-feed market smoke	BOUNDED	Confirms the repo-native acquisition/freeze/refresh path on a two-series sample without upgrading it into authority-bearing market proof.
Enterprise benchmark gate	BLOCKED_MISSING_INPUTS	The gate remains open until the named missing exports, query catalog, and truth labels exist.
DB breadth	SQLite supported; Timescale INCONCLUSIVE	Current retained evidence does not justify broader Timescale-backed equivalence claims.

Surface Status Snapshot

Status is reported per evidence surface, not by rhetorical confidence. Each row below reflects the current retained repo truth.

Surface	Status	Proved now	Boundary and evidence
Package / install	SUPPORTED	The repo builds, installs, imports, and clears the retained pytest alignment run.	Anchored by the March 21 operations logs under proofs/artifacts/operations/.
OHLCV compression	SUPPORTED	19.19x on the carried Wave-1 controlled corpus.	Bounded to proofs/artifacts/2026-02-21_ft_wave1_final/ft_ohlcv_benchmark.json.
Tick compression	SUPPORTED	20.57x on the carried Wave-1 controlled corpus.	Bounded to proofs/artifacts/2026-02-21_ft_wave1_final/ft_tick_benchmark.json.
Pattern search	SUPPORTED	Mean P@10 = 0.90 on the carried canonical-template workload.	Bounded to the controlled bundle; the delayed-feed smoke remains NEEDS_LABELS.
Query latency	SUPPORTED	Local query latency p95 = 0.0567 ms on the carried benchmark workload.	Current retained evidence is local-bundle truth, not a broad DB displacement claim.
Delayed-feed smoke	BOUNDED	The repo-native market data path is real on SPY and AAPL.	Bounded by proofs/reruns/2026-03-19_alpaca_demo_smoke/; not authority-bearing market proof.
Enterprise benchmark	BLOCKED	A named blocker packet exists instead of a false pass narrative.	The governing blocker lives at proofs/reruns/2026-03-21_phase06_contract_freeze_attempt_v3/missing_inputs_packet.json.

Throughput

Compression, query latency, and pattern-search headlines remain bounded to the retained Wave-1 controlled corpus. They are useful, but they are not the same thing as enterprise-market benchmark closure.

OHLCV compression 19.19x Current retained controlled-corpus compression anchor.	Tick compression 20.57x Current retained controlled-corpus tick compression anchor.
Pattern search mean P@10 0.90 Current retained controlled workload retrieval quality.	Query latency p95 0.0567 ms Current retained local benchmark query latency anchor.

Measure	Locked value	Meaning
OHLCV compression	19.19x	Compression on the carried Wave-1 controlled corpus.
Tick compression	20.57x	Tick-stream compression on the carried Wave-1 controlled corpus.
Max RMSE	0.0	Retained fidelity ceiling on the carried bundle.
Pattern search mean P@10	0.90	Controlled canonical-template retrieval quality.
Query latency	p95 = 0.0567 ms	Local carried-bundle latency, not a broad warehouse benchmark claim.
DB breadth	SQLite supported; Timescale INCONCLUSIVE	Current DB truth remains narrower than full incumbent-comparison narratives.

Public Market Workbooks

This repo exposes repo-native observability hooks, but it does not currently promote an external public workbook as an authority-bearing front-door claim. Use the proof artifacts and docs below as the public evidence path.

Role	Surface	Meaning
Lane-local Comet adapter	python/zpe_finance/comet_logging.py	Defines the workstream-local Comet project defaults and logging hooks.
Dual-tracker observability flow	python/zpe_finance/tracking.py	Carries the verify-or-create tracking pattern for classic Comet and Opik.
Observability design note	docs/INTEGRATION_PATTERN.md	Explains how FT tracking aligns with the broader repo family pattern.
Public evidence path	proofs/artifacts/operations/	Current public-facing proof path for install/build/test reality in this repo.

Go Next

If you need to...	Open this
Audit the current front-door truth quickly	docs/AUDITOR_PLAYBOOK.md
Understand the runtime, helper, scripts, and proof layout	docs/ARCHITECTURE.md
Read the bounded public audit rules and non-claims	docs/PUBLIC_AUDIT_LIMITS.md
Inspect proof posture and claim-by-claim evidence	proofs/FINAL_STATUS.md and proofs/CONSOLIDATED_PROOF_REPORT.md
Inspect the current enterprise blocker directly	proofs/reruns/2026-03-21_phase06_contract_freeze_attempt_v3/missing_inputs_packet.json
Understand integration and observability details	docs/INTEGRATION_PATTERN.md

Observability: Comet dashboard (public)

Repo Shape

Field	Value
Proof Anchors	5
Modality Lanes	2
Authority Source	proofs/FINAL_STATUS.md

Modality Lanes counts the two retained financial data lanes called out in the current proof surface: OHLCV bars and tick streams.

Directory Map

Area	Purpose
README.md, CHANGELOG.md, CONTRIBUTING.md, SECURITY.md, CITATION.cff, LICENSE	Root governance, release metadata, and legal terms
python/, core/, scripts/	Installable package, optional helper, and repo-local runners
docs/	Audit, routing, architecture, support, and contract docs
proofs/	Controlled bundle, reruns, proof summaries, and historical boundaries

Open Risks (Non-Blocking)

• The optional native helper is still a deliberate opt-in path; the truthful default quick verify may remain on python-fallback.
• The delayed-feed March smoke is real but bounded to a two-series sample and does not close the enterprise benchmark.
• Timescale-backed DB equivalence remains unresolved.
• The live public repo can tell the truth without being public-release-ready; do not collapse those states.
• Historical lineage artifacts may preserve machine-local traces and should be read as evidence, not as live front-door instructions.

Who This Is For

Ideal first buyer	Market-data infrastructure team or quant research platform
Pain	Delayed-feed archives and replay pipelines need deterministic fidelity guarantees, not just generic compression
Deployment	SDK — Python package with optional Rust native helper
Family position	Secondary product candidate in the Zer0pa deterministic encoding family. Not the lead commercial front door

Contributing, Security, Support

Contribution workflow: CONTRIBUTING.md	Security policy and reporting: SECURITY.md	Support routing: docs/SUPPORT.md
Frequently asked questions: docs/FAQ.md	Autonomous agents and AI systems are subject to Section 6 of the Zer0pa SAL v6.2.

Portfolio Ecosystem

Repo	Package	Wedge
ZPE-IMC	zpe-multimodal	Cross-modal codec. Image, video, audio, point cloud.
ZPE-IoT	zpe-iot	Sensor codec. Telemetry retrieval. Edge deployment.
ZPE-XR	zpe-xr	XR motion codec. Sub-mm fidelity. CPU-native.
ZPE-Robotics	zpe-robotics	Robot trajectory codec. Real robot benchmark.
ZPE-Geo	zpe-geo	Geospatial trajectory codec. Query-aware retrieval.
ZPE-FT	zpe-ft	Market-data codec. OHLCV, ticks, pattern search.
ZPE-Ink	zpe-ink	Vector stroke codec. Structured drawing tiers.
ZPE-Neuro	zpe-neuro	EEG codec. Deterministic primitives. Retrieval harness.
ZPE-Mocap	zpe-mocap	Motion capture codec. Verification-led surface.
ZPE-Prosody	zpe-prosody	Prosody codec. Pitch, rhythm, stress.
ZPE-Bio	zpe-bio	ECG and EEG codec. Deterministic round-trip.