1 第 1 層 · 聚合
來自四大信號來源：經典策略 · 因子 · 機器學習 · AI 假設的信號。
Bring classic strategies, factors, ML models, and AI-assisted hypotheses into an inspectable candidate set. Candidate generation does not establish validity.
AI-Native Quant Research
StratCraft
Move from hypothesis to data, candidates, experiments, reproducible evidence, and the next research decision.
AI guides the workflow. You remain in control.
AI-Native Quant Research · human-controlled
Research evidence · evaluation and transparent baselines
Illustrative interface, not a performance claim500–1000×
workload-specific benchmarks
2000+/sess
reviewed candidates
7
LLM 供應商
1000+
governed research set
Research Capabilities in Context
Product maturity is measured by evidence and governed decisions, not candidate volume.
0
AI helps frame, build, and explain research
0
Users approve operations and next decisions
0
Community execution, Artifacts, evidence, and lineage
0
Community, hosted inference, and Commercial boundaries
The Research Capability Chain
AI connects research work while the user controls operations and decisions.
2 第 2 層 · 速度
Local C++23 execution, with workload-specific evidence.
Run approved backtests locally and preserve inputs, Artifacts, lineage, and results. Performance claims remain benchmark-specific.
Python
(vectorbt)1×
(vectorbt)1×
Rust
(NautilusTrader)42×
(NautilusTrader)42×
StratCraft
C++23784×
C++23784×
3 第 3 層 · 組合
Transparent evaluation. Governance before promotion.
Compare candidates against appropriate baselines and retain acceptance or rejection evidence. Advanced fusion remains separately admitted.
∞ 全本地 · 無鎖定
全本地，無鎖定
Community AI Studio supports loginless direct BYOK or a supported local model. Optional hosted inference is a separate, disclosed boundary.
Windows
macOS
Linux
Why Research Continuity Matters
Connected evidence and decisions matter more than generating a larger candidate count.
傳統方法
StratCraft Research Workflow
信號來源
手動編寫 3-5 個策略
pages.quantnexus.scaleComparison.row1Quantnexus
回測吞吐量
Disconnected tools and incomplete run context
pages.quantnexus.scaleComparison.row2Quantnexus
組合建構
選出最佳策略單獨運行
pages.quantnexus.scaleComparison.row3Quantnexus
結構化方法
集中風險, 一個策略失敗，全盤皆輸
pages.quantnexus.scaleComparison.row4Quantnexus
A larger search space creates more opportunities for false discoveries. StratCraft treats evidence, rejection, and research memory as first-class product responsibilities.
三層流水線
1
第一層: 從四大來源聚合
匯集經典策略（開源函式庫、已發表系統、TradingView、你自己的成果）、量化因子、機器學習模型, 以及 LLM 生成的假設。一個想法變成一整個候選群體, 而非單一手寫策略。
2
第二層: 回測整個工廠
The local C++23 engine runs approved backtests and retains inspectable results. Named data routes and performance evidence are stated per workload.
3
第三層: 機構級統計組合
Compare surviving candidates with transparent baselines. Community includes equal-weight combination and replay; advanced fusion and decision policies are separately admitted.