1. Project Introduction
StockSeer is an intelligent analysis system for A-shares that combines quantitative backtesting, knowledge graphs and multi-analyst LLM debate to automatically generate structured price prediction reports. The system follows the core philosophy of "data first, quantitative priority", ensuring every conclusion is traceable to data.
Online Access
- Project site: https://seer.astock.com.cn/
Core Principles
- Data first: quantitative analysis accounts for 70% of the final score, LLM text judgment only 30%, avoiding pure subjective guesses.
- Full validation: all algorithms are validated on the full stock universe; sampling is prohibited.
- Reproducibility: every analysis result can be fully reproduced with the same inputs, eliminating black-box operations.
2. Core Capabilities
- Quantitative analysis: covers momentum, price-volume and sentiment factor classes; all algorithms are backtested on full data and output multi-dimensional quantitative metrics
- Multi-source data collection: aggregates multiple news and announcement sources plus search API supplementation, gathering dozens of relevant items per analysis
- Knowledge graph: builds relation graphs between stocks and events with visual exploration
- Multi-analyst debate: multiple virtual analysts from different perspectives conduct multiple rounds of structured debate; conclusions are weighted-fused with quantitative scores to avoid single-view bias
- Automated reports: automatically generates standardized price prediction reports with history archiving and comparison
3. Uptime and Stability
The system has been running stably for 128 days since 2026-04-25 (as of 2026-08-31). It automatically runs full analysis flows twice per weekday, with 113 completed analysis tasks and service availability above 99%.
3.1 Key Stability Metrics
| Metric | Value | Period |
|---|---|---|
| Continuous running days | 128 days | 2026-04-25 ~ 2026-08-31 |
| Analysis tasks completed | 113 | cron records |
| Average daily frequency | 1.7/day | 2/day on weekdays |
| Success rate | >99% | log-based |
| Longest continuous run | 25+ days | since 2026-08-03 |
3.2 Stability Safeguards
- Health pre-checks: automatically verifies service reachability before each run; aborts on failure to avoid wasted runs
- Holiday skipping: automatically identifies non-trading days
- Anti-duplication: the same task never runs twice in one day
- Dual-path logging: key logs double-written for zero loss
- Self-healing: automatic cleanup on abnormal service startup
- Timeout protection: multi-layer timeouts with parsing-failure fallbacks, so tasks always produce results
- Concurrency control: file-write locks prevent race conditions
3.3 Version Evolution
| Version | Date | Key improvement |
|---|---|---|
| v1.0 | 2026-04-25 | Basic analysis pipeline launched |
| v1.1 | 2026-04-27 | Debate engine concurrency optimization |
| v1.2 | 2026-04-27 | Frontend timeout fixes, 151 unit tests passing |
| v1.3 | 2026-04-27 | Layout refactor, graph visualization embedded |
| v1.4 | 2026-04-26 | News collection sources expanded |
| v1.5 | 2026-06-09 | Quantitative engine integrated |
| v1.6 | 2026-06-10 | Health pre-checks, holiday skipping, dual-path logs |
4. System Runtime Architecture
4.1 Architecture Source
The following architecture information was generated by Archify from real source-code analysis of the project repository. Every module maps to a real source file, giving the diagram a verifiable, explainable code-to-diagram correspondence.
- Full interactive architecture diagram (Archify HTML output): Open architecture diagram
Interactive diagram notes
Open the link above to view the full interactive architecture diagram, supporting node search, route tracing, semantic views, theme switching, zoom/pan and presentation playback.
4.2 Architecture Overview
4.3 Core Components
| Component | Type | Role | Source files |
|---|---|---|---|
| Browser / Frontend | External | Vue 3 Web UI | frontend/src/App.vue |
| Vue 3 Frontend | Frontend | pages + router | frontend/src/App.vue |
| Flask Backend API | Backend | REST + task runner | backend/run.py, backend/app/api/stock.py |
| Price Monitor Service | Backend | trade signals | backend/app/services/trade_signal.py |
| Data Sync Service | Backend | market / news / fundamentals | backend/app/services/stock_data_collector.py |
| GitHub Source | External | upstream repository | — |
| Local Data & Snapshot | Database | history data + config | stock_analysis/data/sh605118.json, config.yml |
| Scheduler / Cron Jobs | Backend | long-running task state | backend/app/models/task.py |
4.4 Data Flow and Connections
| Connection | Direction | Description |
|---|---|---|
| HTTPS | Browser / Frontend → Vue 3 Frontend | User entry point |
| REST API | Vue 3 Frontend → Flask Backend API | Frontend calls backend APIs |
| data requests | Flask Backend API → Price Monitor Service | Issues data requests |
| snapshot | Price Monitor Service → Local Data & Snapshot | Writes snapshots |
| trigger sync | Flask Backend API → Data Sync Service | Triggers data sync |
| fetch market data | Data Sync Service → GitHub Source | Fetches upstream market data |
| upstream feed | GitHub Source → Local Data & Snapshot | Upstream data flows in locally |
| heartbeat | Scheduler → Flask Backend API | Periodic health probe |
| market checks | Scheduler → Price Monitor Service | Periodic market checks |
4.5 Three Operational Perspectives
User Entry
- Vue pages serve the interactive UI over HTTPS.
- Requests reach the Flask API and the task runner.
Data Path
- The collector fetches market, news, and fundamentals.
- History data and snapshots live in local storage.
Automation
- Task orchestration tracks long-running pipeline jobs.
- Signals and sync keep snapshots aligned with upstream data.
The diagram includes two views: Primary runtime flow (user-facing requests pass through the Vue frontend, Flask API, data services, and local snapshots) and Data sync path (upstream data flows through the collector into local snapshots, tracked by task orchestration).
5. Performance
A full analysis flow for a single stock averages about 15~26 minutes, varying with data volume and local LLM inference speed. Resource usage is healthy: frontend/backend resident memory is around 100 MB, CPU is near-idle, and graph database and LLM services use resources on demand.
6. Security & Compliance
- Data privacy: LLM runs locally, sensitive data never leaves the intranet
- Data integrity: quantitative backtesting must use the full stock universe, sampling prohibited
- Auditability: dual-path logs record every execution for post-hoc tracing
- Version control: code and configuration under Git management
This document is the public introduction of the StockSeer system, focusing on core capabilities and stability. Internal architecture, strategy parameters and backtest details are internal research materials and are not displayed here.