What does this skill do, and when should you use it?
This is one of 14 skills bundled in olaxbt/ai-market-maker, an open-source, hedge-fund-style agentic trading system for crypto. Located at src/agents/operator/2.2_statistical_alpha_engine/, it computes cross-sectional rankings and z-scores from universe-wide OHLCV data plus a factor cache covering momentum, value, volatility and on-chain metrics. Its analyze() call returns a Tier-0 contract with rank, z-score, alpha signal and factor confluence. It depends on the parent repo's universe config and factor pipeline, so it cannot run standalone. It suits quant-minded users already running the AIMM stack who need a statistical research-desk signal.
- Reads universe-wide OHLCV data and a factor-model cache (momentum, value, volatility, on-chain)
- Exposes analyze(ticker, market_data) returning a Tier-0 contract with cross-sectional ranking
- Answers direct queries such as 'How does BTC rank in the universe?' or 'What's the cross-sectional z-score of ETH?'
- Serves a query endpoint /statistical_alpha_engine?ticker=BTC/USDT returning rank, z-score, alpha signal and factor confluence
- A quant researcher who needs a fast relative-strength rank for a specific coin across a multi-asset universe
- The Statistical Alpha research desk inside AIMM, which needs factor-level inputs before signal synthesis
- A developer checking whether momentum, value, volatility and on-chain factors align (confluence) for one asset
- A backtesting workflow that screens relatively strong or weak assets by z-score
- Users who want to run the skill standalone outside AIMM — it hard-depends on the parent repo's universe config (src/config/default_universe.py) and factor pipeline
- Traders seeking a proven, ready-to-deploy strategy — the README warns one profitable backtest is not proof of edge, requiring multi-regime, out-of-sample validation, and no standalone accuracy or test evidence exists for this skill
- Commercial SaaS operators unwilling to meet AGPL-3.0 source-offer obligations when running a modified version as a network service
How do you install this skill?
- Alpha signals and Factor_Confluence rules are unverified by execution; do not use for real trading decisions — the README itself warns a profitable backtest is not proof of edge.
- The skill does not declare behavior boundaries on missing data or universe-size changes; failure feedback on abnormal input is unknown.
- No Chinese documentation; on-chain/factor data depends on overseas crypto data pipelines with unverified mainland-China reachability.
- Skill-level versioning and maintenance responsibility are not declared; track repo changes yourself for upgrade compatibility.
- Publisher is not verified by the FollowSkills registry — identity is unknown (not a deduction by itself, but note it).
- Shell / CLI
- Network access
- Local filesystem
Python 3.11+uvTA-Libparent ai-market-maker repo (src/config/default_universe.py, factor pipeline)
The skill ships inside the parent repo, not as a standalone package. Install via:
OpenClaw
claw install https://github.com/olaxbt/ai-market-makerManual
git clone https://github.com/olaxbt/ai-market-maker.git
cd ai-market-maker
conda install -y ta-lib -c conda-forge
uv sync --extra devThe skill file lives at src/agents/operator/2.2_statistical_alpha_engine/SKILL.md.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- How does BTC rank in the universe?
- What's the cross-sectional z-score of ETH?
- /statistical_alpha_engine?ticker=BTC/USDT
With the parent repo deployed, call the query interface: /statistical_alpha_engine?ticker=BTC/USDT to get cross-sectional rank, z-score, alpha signal and factor confluence; or ask directly in conversation, e.g. 'How does BTC rank in the universe?' or 'What's the cross-sectional z-score of ETH?'. SKILL.md documents no standalone CLI command; the exact invocation path depends on the parent repo's API/agent flow.
What are this skill's strengths and limitations?
- Cross-sectional rank plus z-score is a standard quant method with structured (Tier-0 contract) output, easy to feed downstream
- Factor coverage spans momentum, value, volatility and on-chain metrics, not price action alone
- Part of an actively presented open-source system under AGPL-3.0 with auditable code
- Hard dependency on the parent repo (universe config at src/config/default_universe.py and the factor pipeline) — unusable outside AIMM
- The README itself warns that one profitable backtest is not proof of edge; multi-regime, out-of-sample validation is required
- No standalone accuracy or test evidence for this skill is provided in the source material
- AGPL-3.0 imposes source-offer obligations if you run a modified version as a network service — evaluate before commercial SaaS use
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Statistical Alpha Engine (Skill 2.2) this page | 47 · Use with care | ★ 2.1k | 2mo ago | AGPL-3.0 |
| Pro Bias Analyst | 44 · Not recommended | ★ 2.1k | 2mo ago | AGPL-3.0 |
| Pattern Recognition Bot (2.1) | 42 · Not recommended | ★ 2.1k | 2mo ago | AGPL-3.0 |
| WQ Alpha Research Skill | 51 · Use with care | ★ 409 | 3mo ago | — |
| Liquidity Order Flow Analyst (4.2) | 47 · Use with care | ★ 2.1k | 2mo ago | AGPL-3.0 |
Within the same AIMM system it sits alongside the Technical TA Engine (patterns and indicators) and Sentiment & Narrative (news and hype) desks; this skill is the factor/cross-sectional statistics specialist.
How did FollowSkills review this skill?
Read-only statistical analysis with no funds movement, secrets, or external writes; no red-line risk found. SKILL.md declares data sources and dependency paths. However, data-flow details (origin of on-chain factors) are undisclosed, there is no financial-risk disclaimer or confirmation mechanism, and LLM-upstream transparency is limited. Deducted for incomplete data-flow disclosure and missing scoping/confirmation/rollback.
persona.md provides a Tier-0 output contract, reasoning rules (z≥|2.0|, decile classification) and two few-shot examples; the happy path within the framework is plausible. But no skill-specific test coverage is shown (provided tests do not touch 2.2's key paths), failure feedback on abnormal input is absent, and Factor_Confluence rules are loosely specified (fixed 95 in the example). Deducted for insufficient key-path reproduction evidence and thin error handling.
Scenario is clear (cross-sectional rank, z-score queries for BTC/ETH); trigger semantics reasonably precise. But capability boundaries (universe size, behavior on missing data, non-fit ranges) are undeclared, there is no Chinese-language support, and environment-fit evidence is limited given the crypto data pipeline. Deducted for missing boundaries and non-fit declaration.
SKILL.md is well-structured with dependencies and query interface declared; persona supplies schema and few-shots; repo has AGPL-3.0, versioned pyproject, CI and maintenance signals. But the skill itself lacks version/changelog, known-limitation disclosure, FAQ, and its governance relation to the other 13 SKILL.md files is unexplained. Deducted for skill-level versioning and limitation disclosure gaps.
Output contract is structured JSON (tier0/v1) with rank, z-score and alpha signal, directly consumable downstream; real marginal value as part of the pipeline. But static review cannot verify output correctness or signal validity, the README itself warns a profitable backtest is not proof of edge, and no comparative-benefit evidence exists. Deducted for unverified results and weak comparative-benefit evidence.
Repo provides CI workflows, a pytest suite, and pinned data with MANIFEST hashes — auditable primary material. But the shown tests do not cover the 2.2 alpha engine's key paths, few-shot outputs are author-authored, and there is no third-party execution evidence or cross-source corroboration. Deducted for thin reproducible coverage and unseparated claims.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →