tick-stock-panel

TSP (tick-stock-panel) is a self-hosted quantitative workbench for the A-share market, designed to provide a one-stop solution for stock screening, monitoring, backtesting, and more. The project emphasizes zero-ops, deploying via a single Docker container, with all data stored in local Parquet files. Core features include: - **Capability routing matrix**: 6 categories of datasets are routed independently by source capability, allowing data sources to be swapped at any time without affecting indicators or backtest conventions. - **Stock screening engine**: Built-in 25 strategies, supports custom signals and AI-generated strategies, and can scan the entire A-share market at millisecond speed. - **Indicator pipeline**: Computes 68 columns of indicators and signals based on Polars, unifying data conventions. - **Backtest research**: Supports factor backtesting, strategy backtesting, and minute-level strategy backtesting, accounting for factors such as T+1, fees, and slippage, and provides factor attribution analysis. - **Factor platform**: Supports DSL-based custom factors and validation combinations, with two-way linkage to strategies. - **Factor mining**: Out-of-sample search for multi-factor combinations, explicitly published, never automatically deployed. - **Market environment**: Identifies 6 stages of the sentiment cycle and provides concept/industry mainline rankings. - **Anomaly monitoring**: Covers three types of anomalies—auction, intraday, and deviation—with support for real-time pop-ups, voice announcements, and Feishu push notifications. - **Monitoring center**: Four types of rules (strategy, individual stock signals, price, anomalies) support AND/OR combinations for real-time monitoring. - **Individual stock analysis**: Provides 9 types of key price levels and AI four-dimensional analysis (technical, fundamental, financial, news). - **Consecutive limit-up tiers**: Tallies consecutive limit-up levels, with support for concept rotation and post-market AI review. - **Data extension**: Pluggable data sources, with support for extended fields and daily historical backfill. - **AI chat assistant**: 18 read-only tools covering the capabilities of all site pages, with support for character-by-character streaming output and tool-call trace cards, as a fully decoupled extension module. The technical architecture uses a layered design: data source layer (pluggable), capability routing layer, storage layer (Parquet, DuckDB), computation layer (Polars), research layer (factor engine, backtest engine), application layer (FastAPI), presentation layer (React 18). It supports multiple deployment methods, including ready-made GHCR images, Docker Compose, local AI-assisted deployment, and Dev mode. The project is for learning and research purposes only and does not constitute investment advice.