An evidence-driven AI quantitative workstation for China's A-share market.
This repository presents the public product architecture, safety model, and delivery principles of Archē Quant. Production source code, trading credentials, proprietary market data, and broker integrations remain private.
| Document | What it shows |
|---|---|
| Product tour | The end-to-end user workflow and visible outcomes |
| System architecture | Data, research, agent, execution, and observability boundaries |
| Safety model | Authorization, market-data, broker, and reconciliation gates |
| Roadmap | What is public now and what may be released next |
| Security policy | Safe disclosure and public-data boundary |
A selected strategy exposes its trigger description, hit count, ranking, market metrics, and signal labels on the real product page.
The analysis workspace combines daily candles, moving averages, volume, support/resistance, channel levels, pivot points, and volume-profile levels.
Simulation notice: this screenshot uses a simulated account. All asset, position, profit/loss, and order values are non-production demonstration data and do not represent real funds or live brokerage authority.
The trading workspace exposes account freshness, allocation, equity history, strategy runtime, signal queue, risk-gate state, orders, fills, and reconciliation evidence.
Screenshots intentionally exclude account identifiers, credentials, broker configuration, and internal endpoints.## Product loop
AI research → data screening → backtest → live monitoring
→ risk gate → approved intent → broker → reconciliation
Archē Quant is designed as a complete decision and execution workspace rather than an isolated strategy script.
- Visible evidence: a feature is not complete until the real page, data path, and user interaction agree.
- Data truth: incomplete market snapshots do not replace the last validated full snapshot.
- Explicit authority: strategy signals never become orders without risk and authorization gates.
- External closure: request acceptance is not treated as a fill, cancellation, or final state.
- Fail-closed operation: unknown permissions, stale data, broker ambiguity, or reconciliation mismatch halt consequential writes.
| Layer | Stack |
|---|---|
| Frontend | React, TypeScript, Tailwind CSS, TanStack Query, ECharts |
| Backend | FastAPI, Polars, DuckDB, vectorbt |
| Agent runtime | Tool calling with explicit capability boundaries |
| Data | Parquet, market-data providers, QMT |
| Delivery | Containerized services, health probes, durable event records |
This showcase intentionally excludes account identifiers, credentials, private strategies, proprietary market data, broker endpoints, operational logs, and code that could enable unauthorized trading.
Active private development. The documents in this repository describe architecture and engineering practice; they are not claims that every production workflow has completed live external acceptance.
Signal is not order. Request success is not terminal state.
Documentation and image assets are available under CC BY 4.0.