neogenesislab/quant-v11-ensemble-6alpha-specs-2026

DOI


This dataset is citable via DataCite DOI `10.5281/zenodo.20018487` (Zenodo record).

Cite as:

@dataset{neogenesis_20018487,
  author       = {Heo, Yesol and Neo Genesis Lab},
  title        = {Quant v11 Ensemble 6-Alpha Specs & Risk Engineering 2026},
  year         = 2026,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.20018487},
  url          = {https://doi.org/10.5281/zenodo.20018487}
}

Quant v11 Ensemble - 6-Alpha Specs & Risk Engineering 2026

A reference systems-engineering corpus for a 6-alpha cryptocurrency perpetual futures ensemble bot. 375 parsed sections from 19 canonical source files: 67 design, 46 risk-layer, 115 alpha-spec, 71 expert-report, 42 research, 34 roadmap sections.

Released by [Neo Genesis](https://neogenesis.app) (Wikidata Q139569680).

Educational disclaimer (read first)

• —Not investment advice. This dataset is published for research and engineering education.
• —PAPER mode. The reference v11 ensemble runs in PAPER mode (no live capital) until a 14-day Sharpe >= 1.2 + Deflated Sharpe Ratio >= 0.5 gate clears on at least one alpha.
• —Edge decay. Quant alpha edges decay from the moment they are documented publicly. Anything in this corpus must be re-validated by the reader against fresh data, real fees, real slippage, and real funding before any capital is risked.
• —Survivorship + lookahead. v6-v10 of this same bot produced "127k wins / 0 losses" backtests that turned into a -<illustrative-loss> drain on 5 days of live trading. The v11 redesign exists because the old backtests were wrong. Treat any backtest, including the ones implied here, with that history in mind.

Why this dataset is unique

Public quant ML / RL datasets (price candles, order books, returns) are common. Full systems-engineering specs of a working multi-alpha derivatives bot - alphas + risk + experts + research + roadmap as one coherent system - are rare publicly. What is in here that you usually have to reconstruct yourself:

1. 1.6 low-correlation alpha specifications (A1..A6) with theory background, entry/exit logic, paper-mode KPIs, expected daily contribution, and stop-go criteria.
2. 2.9-layer kill switch (Layer 1 = per-position SL, ..., Layer 9 = funding-spike guard) with explicit thresholds, not slogans.
3. 3.6 cross-discipline expert validation reports - math boundary (Kelly + ruin probability), HFT / market-making, statistical arbitrage, risk survival, ML / RL, event alpha. Each report cross-checks the alphas independently.
4. 4.External benchmark research (2026-04-24) on what public benchmarks exist for liquidation-cascade alphas, OU-mean-reversion, funding arbitrage, and the absence of public benchmarks for several of these.
5. 5.Backtest engine decision - why the v11 ensemble picks nautilus_trader (primary) + hftbacktest (microstructure-sensitive A1 / A6) + vectorbt pro (cross-validation), with the 8 anti-patterns the v6-v10 backtests fell into.

This is a real design used by a single-operator quant stack (operator publicly attested via Wikidata Q139569708). It is not a paper proposal.

Dataset summary

• —375 sections in data/sections.jsonl. Each row is one parsed heading-section.
• —19 source files indexed in data/source_index.json:
• —Design (67 sections): MASTER_DESIGN.md, INDEX.md, CURRENT_CODE_AUDIT.md, backtest-v2-engine-decision.md.
• —Risk layers (46 sections): RISK_KILLSWITCH.md - 9-layer defense.
• —Alpha specs (115 sections): A1 Liquidation Cascade, A2 Mean Reversion OU, A3 Extreme Funding, A4 Macro Event, A5 Funding / Basis Harvest, A6 Alt Market-Making.
• —Expert reports (71 sections): math boundary, HFT / MM, stat arb, risk survival, ML / RL, event alpha.
• —Research (42 sections): external benchmark validation 2026-04-24.
• —Roadmap (34 sections): Phase -1 / 0 / 1 progression.

Schema

data/sections.jsonl - one JSON record per line:

Quick start

from datasets import load_dataset

ds = load_dataset("neogenesislab/quant-v11-ensemble-6alpha-specs-2026", "sections", split="train")
print(len(ds), "sections")

# Get all alpha specs
alphas = ds.filter(lambda r: r["doc_type"] == "alpha_spec")
print(f"{len(alphas)} alpha-spec sections")

# Filter to one specific alpha (A1 Liquidation Cascade)
a1 = ds.filter(lambda r: r["alpha_id"] == "A1")
for row in a1:
    print(row["section_title"])

# Get all risk-layer sections that reference Layer 3 (correlation killer)
l3 = ds.filter(
    lambda r: r["doc_type"] == "risk_layer" and r["kill_switch_layer"] == 3
)

# Get all expert reports
experts = ds.filter(lambda r: r["doc_type"] == "expert_report")

Suggested research applications

1. 1.Multi-alpha ensemble design - train portfolio allocators or meta-classifiers to gate alpha activation under regime conditions.
2. 2.Kill-switch reasoning - supervise classifiers that decide when to halt trading based on correlation, drawdown, fee burn, or stablecoin depeg signals.
3. 3.Backtest anti-pattern detection - the backtest-decision doc enumerates 8 specific anti-patterns (lookahead bias, survivorship, fill-model bugs, missing funding, missing slippage, etc.) - useful as supervision for backtest auditors.
4. 4.Cross-discipline alpha review - 6 expert reports give you parallel views of the same 6 alphas. Useful as a small benchmark for "do LLM critics agree with domain experts."
5. 5.Single-operator quant ops - rare reference for what one person actually has to specify (alphas + risk + telemetry + roadmap + experts) to ship a derivatives bot honestly.

Operating status

• —Mode: PAPER (no live capital deployed)
• —Graduation gate: 14-day Sharpe >= 1.2 AND Deflated Sharpe Ratio >= 0.5 on the active alpha
• —Max leverage hard cap: 5x (Kelly/3 safety factor)
• —Safety layers: 9 (per-position SL, daily/weekly/monthly DD, correlation killer, fee-budget freeze, halt persistence, order-rate cap, stablecoin depeg guard, funding-spike guard)

Anonymization

Each emitted string is re-tested with all redaction regexes; publish aborts on leak.

Provenance

• —Source: Neo Genesis private SSOT (auto-trading/docs/v11-ensemble/*).
• —Curator: Neo Genesis Lab (neogenesislab HuggingFace org).
• —Wikidata: Q139569680 (Neo Genesis), Q139569708 (Yesol Heo, founder).
• —Related releases by the same operator:
• —`korean-rag-ssot-golden-50`
• —`ethicaai-mixed-safe-evidence`
• —`whylab-gemini-2-5-docker-validation`
• —`sbu-pseo-effects-2026-04`
• —`cross-agent-review-queue-2026`
• —`korean-llm-citation-baseline-2026`
• —`sora-multi-device-orchestration-2026`

Citation

@misc{neogenesis_quant_v11_2026,
  title  = {Quant v11 Ensemble 6-Alpha Specs and Risk Engineering 2026: A reference systems-engineering corpus for a multi-alpha cryptocurrency perpetual futures bot},
  author = {Neo Genesis Lab},
  year   = {2026},
  url    = {https://huggingface.co/datasets/neogenesislab/quant-v11-ensemble-6alpha-specs-2026},
  note   = {6 alphas, 9-layer kill switch, 6 expert validations, external benchmark research; published in PAPER mode pending 14-day Sharpe >= 1.2 + DSR >= 0.5 graduation gate}
}

License

CC-BY-4.0 - free for research and commercial use with attribution to Neo Genesis Lab.

한국어 요약

Quant v11 Ensemble 6-Alpha Specs & Risk Engineering 2026 은 Binance 선물에서 일 평균 0.6~1.0% 누적 수익률을 목표로 하는 6-알파 앙상블 봇의 시스템 엔지니어링 SSOT 코퍼스다.

교육용 면책 (먼저 읽기)

• —투자 자문 아님. 본 데이터셋은 연구 / 엔지니어링 교육 목적으로 공개된다.
• —PAPER 모드. 본 v11 reference 앙상블은 14일 Sharpe >= 1.2 + Deflated Sharpe Ratio >= 0.5 게이트를 통과할 때까지 PAPER 모드에서만 운영된다 (실자본 0).
• —알파 감쇠. 퀀트 알파 edge 는 공개 시점부터 감쇠한다. 본 코퍼스의 어떤 내용도 독자가 자신의 데이터 / 수수료 / 슬리피지 / funding 가정으로 재검증하지 않으면 그대로 사용해서는 안 된다.
• —생존편향 + 룩어헤드. 동일 봇의 v6-v10 은 "127k 승 / 0 패" 백테스트를 만들어 냈고 라이브에서 5일 만에 -<illustrative-loss> 드레인이 발생했다. v11 의 존재 자체가 옛 백테스트가 틀렸기 때문이다. 본 코퍼스가 암시하는 어떤 백테스트도 그 이력 하에서 읽어야 한다.

다루는 것

• —6개 저상관 알파 스펙 (A1 Liquidation Cascade, A2 Mean Reversion OU, A3 Extreme Funding, A4 Macro Event, A5 Funding / Basis Harvest, A6 Alt Market-Making) - 이론 / 진입출구 / 페이퍼 KPI / 일일 기여 / stop-go 기준.
• —9-Layer Kill Switch - L1 포지션 SL, ..., L9 funding-spike guard. 슬로건이 아니라 임계값으로.
• —6명 전문가 교차 검증 - 수학 (Kelly + 파산확률), HFT / MM, 통계차익, 리스크 생존, ML / RL, 이벤트 알파.
• —외부 벤치마크 리서치 - liquidation-cascade / OU / funding arb 의 공개 벤치마크 부재 공식화.
• —백테스트 엔진 결정 - nautilus_trader 메인 + hftbacktest 보조 + vectorbt 검증, v6-v10 백테스트의 8 anti-pattern.

익명화

디바이스 hostname / 이메일 / 전화 / RRN / API 토큰 / 지갑 주소 / 절대경로 / private IP / 구체적 자본 금액 모두 redaction. 발행 직전 모든 문자열을 redaction regex 로 재검증.

라이선스 CC-BY-4.0 - 인용 시 자유롭게 사용 가능.

Citation

@dataset{neogenesislab_quant_v11_ensemble_6alpha_specs_2026_2026,
  author       = {Yesol Heo and Neo Genesis Lab},
  title        = {Quant v11 Ensemble 6-Alpha Specifications 2026},
  year         = 2026,
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/neogenesislab/quant-v11-ensemble-6alpha-specs-2026},
  note         = {Wikidata Q139569680, Q139569708; license CC-BY-4.0}
}

Citation File Format

GitHub, Zenodo, and other tooling can read the following CFF block to provide one-click citation export (BibTeX, APA, RIS, etc.). The CFF specification is v1.2.0.

cff-version: 1.2.0
message: "If you use this dataset, please cite it as below."
title: "Quant v11 Ensemble 6-Alpha Specs & Risk Engineering 2026"
type: dataset
authors:
  - family-names: "Heo"
    given-names: "Yesol"
    affiliation: "Neo Genesis Lab"
date-released: "2026-05-03"
license: CC-BY-4.0
url: "https://huggingface.co/datasets/neogenesislab/quant-v11-ensemble-6alpha-specs-2026"
repository: "https://huggingface.co/datasets/neogenesislab/quant-v11-ensemble-6alpha-specs-2026"
identifiers:
  - type: doi
    value: "10.5281/zenodo.20018487"
    description: "Zenodo DataCite DOI for this dataset"
  - type: other
    value: "Q139569680"
    description: "Wikidata Q-ID of the publishing organization (Neo Genesis)"
keywords:
  - quantitative-trading
  - kill-switch
  - ensemble
  - liquidation-cascade
  - funding-rate
  - mean-reversion
  - binance
  - neo-genesis
  - risk-engineering
preferred-citation:
  type: dataset
  title: "Quant v11 Ensemble 6-Alpha Specs & Risk Engineering 2026"
  authors:
    - family-names: "Heo"
      given-names: "Yesol"
      affiliation: "Neo Genesis Lab"
  doi: "10.5281/zenodo.20018487"
  year: 2026
  publisher:
    name: "Zenodo"
  url: "https://doi.org/10.5281/zenodo.20018487"