What is this?
A natural-language-driven, low-frequency quantitative trading decision system for China's A-share market. It combines a subjective Pitch decision chain, a hardcoded factor engine, and strict data-based arbitration, with the AI forbidden from predicting individual stock moves.
A natural-language-driven, low-frequency quantitative trading decision system for China's A-share market. It combines a subjective Pitch decision chain, a hardcoded factor engine, and strict data-based arbitration, with the AI forbidden from predicting individual stock moves.Does the AI pick stocks and trade on its own, or just assist?The AI never predicts individual stocks directly — that's a hard constraint. It drives factor mining, auditing, and chats with 'legendary trader' personas, but decisions still flow through an L0 timing to L2 Pitch approval to T+1 execution chain before anything happens.
The AI never predicts individual stocks directly — that's a hard constraint. It drives factor mining, auditing, and chats with 'legendary trader' personas, but decisions still flow through an L0 timing to L2 Pitch approval to T+1 execution chain before anything happens.How do I get it running?New users should use the one-click deployment: grab DSHQuant-v1.0.9-Release.zip from the Release page Assets (not the auto-generated 'Source code (zip)'), then double-click 接入API.cmd and paste your DeepSeek API key — the AI wizard then sets up data sources, config, and verification for you.
New users should use the one-click deployment: grab DSHQuant-v1.0.9-Release.zip from the Release page Assets (not the auto-generated 'Source code (zip)'), then double-click 接入API.cmd and paste your DeepSeek API key — the AI wizard then sets up data sources, config, and verification for you.Do I need a DeepSeek API key to use this?Without a key you only get the hardcoded engine with no AI features at all. Connecting a DeepSeek API key unlocks control, factor mining, audits, and runtime customization. Running from source needs Python 3.10+; the exe needs nothing extra; the optional HARNESS AI console additionally needs Node.js 18+.
Without a key you only get the hardcoded engine with no AI features at all. Connecting a DeepSeek API key unlocks control, factor mining, audits, and runtime customization. Running from source needs Python 3.10+; the exe needs nothing extra; the optional HARNESS AI console additionally needs Node.js 18+.What install pitfall should I watch out for?Do not grab GitHub's auto-generated 'Source code (zip)' — it leaves out the harness/node_modules folder, so the AI console won't work. You need DSHQuant-v1.0.9-Release.zip from the Release Assets, which bundles the full HARNESS runtime.
Do not grab GitHub's auto-generated 'Source code (zip)' — it leaves out the harness/node_modules folder, so the AI console won't work. You need DSHQuant-v1.0.9-Release.zip from the Release Assets, which bundles the full HARNESS runtime.Does it come with stock market data built in?No, the system ships no market data at all — you source it yourself (e.g. from Tushare); the repo only has fetch scripts plus synthetic demo data. Turnover-rate data is missing before 2019, so pre-2019 turnover-based conclusions are invalid.
No, the system ships no market data at all — you source it yourself (e.g. from Tushare); the repo only has fetch scripts plus synthetic demo data. Turnover-rate data is missing before 2019, so pre-2019 turnover-based conclusions are invalid.What do the 123+ factors cover?Factors span turnover, low volatility, reversal, liquidity (Amihud), skew/lottery effects, momentum, low-frequency fundamentals (F-score, SUE, accruals), limit-up/down behavior, institutional activity, industry crowding, and Alpha101-style factors, each locally back-tested on A-share data with point-in-time and T+1 controls.
Factors span turnover, low volatility, reversal, liquidity (Amihud), skew/lottery effects, momentum, low-frequency fundamentals (F-score, SUE, accruals), limit-up/down behavior, institutional activity, industry crowding, and Alpha101-style factors, each locally back-tested on A-share data with point-in-time and T+1 controls.What is the legendary-trader ('牛散') feature?It's a set of 7 personas distilled from well-known real Chinese retail traders you can chat with for stock-picking input; any resulting decision automatically feeds into the five-pool forward-validation pipeline rather than being executed directly.
It's a set of 7 personas distilled from well-known real Chinese retail traders you can chat with for stock-picking input; any resulting decision automatically feeds into the five-pool forward-validation pipeline rather than being executed directly.Can I change the system's behavior without recompiling?Yes — a dynamic Cordis plugin system lets the AI hot-patch runtime behavior through the harness module without rebuilding the codebase.
Yes — a dynamic Cordis plugin system lets the AI hot-patch runtime behavior through the harness module without rebuilding the codebase.What are the 12 Skill files for?They're preloaded method and asset files the AI pulls in on demand, covering factor mining workflows, 'legendary trader' distillation profiles, and system maintenance know-how.
They're preloaded method and asset files the AI pulls in on demand, covering factor mining workflows, 'legendary trader' distillation profiles, and system maintenance know-how.What are similar projects?astra-quant-agent, ai-hedge-fund, QUANTAXIS, daily_stock_analysis, bytex, anti-gambling-trader-tw, hummingbot, awesome-quant
astra-quant-agent, ai-hedge-fund, QUANTAXIS, daily_stock_analysis, bytex, anti-gambling-trader-tw, hummingbot, awesome-quantWhich topics is it in?a-share, risk-management, cordis, backtesting, trading-bot, quant, algorithmic-trading, quantitative-finance, trading, finance, plugin-system, deepseek-harness
a-share, risk-management, cordis, backtesting, trading-bot, quant, algorithmic-trading, quantitative-finance, trading, finance, plugin-system, deepseek-harness