ScottZt/jin-ce-zhi-suan

【金策智算】➡️不靠情绪买卖、不追小道消息，专注用客观数据辅助决策。我们仅提供本地化私有行情数据服务与历史回测工具，帮你把主观想法变成可验证的交易规则，用历史数据检验方法有效性，通过指标监控约束随意操作、控制回撤风险，让交易更有纪律、更落地。本产品为纯量化工具，不荐股、不指导买卖、不预测行情、不承诺收益，所有决策由用户自主判断，只为你提供客观的数据支撑与AI辅助。作者【硅基流码】

📈 金策智算 – 智能投研决策系统

A Python‑based quantitative‑trading platform that structures the whole workflow (strategy generation → risk‑check → execution) around a “Three‑Ministries‑Six‑Departments” model inspired by the Tang‑dynasty bureaucracy. It offers a web dashboard (FastAPI + HTML) for back‑testing, strategy management and, notably, an AI‑driven natural‑language stock‑screening feature that turns Chinese descriptions into executable filter rules.

🎯 What the project does

Feature	What you get
Back‑test engine	Historical data ingestion, signal execution, performance reports, consistency checks.
Multi‑strategy management	Built‑in strategies plus a plug‑in point for user‑defined ones; batch, portfolio and “BLK/TDX” formula import.
Risk‑first workflow	A “gate‑keeping” ministry that can veto trades based on stop‑loss, draw‑down, position limits, etc.
AI condition screening	Write a Chinese rule like “五日内有涨停，涨停后缩量至一半以下，RSI 处于超卖区间”; the system calls a large‑model API, parses the intent, and creates a structured filter + execution rule automatically.
Configurable data sources	AkShare, Tushare, generic REST API, MySQL or PostgreSQL back‑ends.
Web UI	FastAPI server (`server.py`) serves `dashboard.html` where you select strategies, set parameters, launch a single‑ticket back‑test or launch the full‑scale “strategy evolution” board.
Strategy evolution board	Generates, evaluates and iterates strategies (the README calls it Evolution), useful for automated optimisation.

🏗️ Architecture at a glance

Three Ministries (decision chain)
1. 太子院 (Crown Prince) – data validation & distribution.
2. 中书省 (Zhongshu Sheng) – strategy signal generation.
3. 门下省 (Menxia Sheng) – risk review; can veto a trade.
4. 尚书省 (Shangshu Sheng) – execution & settlement.
Six Departments (functional modules)
- 吏部 – strategy registration & lifecycle.
- 户部 – cash, cost and NAV accounting.
- 礼部 – performance reports & ranking.
- 兵部 – order matching & trade management.
- 刑部 – violation logging & risk events.
- 工部 – market data cleaning & indicator calculation.

Code layout

src/
  core/            # ministry orchestration
  ministries/      # department implementations
  strategies/      # built‑in & user strategies
  strategy_intent/ # AI intent parsing
  utils/           # config, data adapters, indicators
dashboard.html    # UI
server.py          # FastAPI entry point
main.py, run_*.py  # CLI helpers for back‑test / live monitoring

🚀 Quick‑start (5‑minute guide)

Install – pip install -r requirements.txt (Python 3.8+).
Add private config – create config.private.json with your Tushare/LLM API keys (see README example).
Run the server – python server.py (the script will auto‑install missing deps if needed).
Open the dashboard (usually http://127.0.0.1:8000).
Pick a strategy, fill stock code, date range and initial cash, then click 单票回测 to see the first result.
For batch runs, BLK/TDX imports or strategy evolution, use the 操作工作台 links in the UI.

🤖 AI‑Condition‑Screening workflow

Type a natural‑language stock‑selection idea in Chinese.
The platform calls the configured LLM (llm_api_key in the private config).
The model returns a JSON‑like structure that separates filter conditions from trade‑execution rules.
The filter runs on the chosen data source to produce a candidate universe; the back‑test engine then simulates trades using the execution rules.

📜 License & commercial terms

Free for personal, academic, or local‑use (non‑commercial) – you can study the code, run back‑tests, and modify it.
Commercial use requires a written license from the author (email: zthx410@163.com). This includes SaaS offerings, selling the software or any profit‑making deployment.
The repository includes a full disclaimer stating the tool is not investment advice and that users bear all trading risk.

🙋‍♀️ Contributing

Fork → branch → PR.
Keep the core scripts runnable.
Document test steps in the PR description.

📌 Bottom line

金策智算 is a fairly complete, open‑source quantitative‑research platform that blends traditional back‑testing with a natural‑language‑to‑strategy pipeline powered by large language models. It is suitable for Python developers or quant hobbyists who want a ready‑made framework to experiment with strategy ideas, risk controls, and AI‑assisted stock‑screening, while still being able to dive into the underlying code.