wzj998 GitHub avatar

fluencer-predict-track

wzj998

A set of AI agent skills/rules for Codex, Claude Code, and Cursor that audits investment predictions from public Chinese social media (Zhihu) by verifying them with authoritative data and producing a Markdown report with backtesting and qualitative analysis.

Stars

32

7-day growth

No data

Forks

11

Open issues

0

License

MIT

Last updated

2026-06-29

AI repository intelligence
FR-AI / ANALYSIS

Why it is worth attention

It systematically quantifies the accuracy of finfluencer predictions using public sources (central banks, exchanges, FRED, etc.) and includes a backtesting engine that simulates portfolio returns, Sharpe ratio, and position management, making prediction validation rigorous and transparent.

Who it is for

  • Investment researchers and analysts
  • Finance content consumers who want to fact-check pundits
  • Quantitative developers building forecast‑audit tools
  • AI assistants users interested in structured reasoning skills

Use cases

  • Verify the accuracy of investment predictions made by well‑known Zhihu authors
  • Generate a comprehensive 'compilation' report with prediction‑level verdicts and statistical summaries
  • Backtest tradable predictions into a simulated portfolio to estimate annualised return and Sharpe ratio
  • Compare the performance of multiple finfluencers across different time spans

Strengths

  • Supports three major AI agent entry points (Codex, Claude Code, Cursor) from a single repository
  • Requires at least 20 prediction samples per target and checks last‑edit timestamps to avoid hindsight bias
  • Validates predictions against authoritative public data sources (central banks, stock exchanges, Yahoo Finance, etc.)
  • Integrates a backtesting module that outputs net‑asset‑value curves, annualised returns, and Sharpe ratios with clear disclaimers about position sizing

Considerations

  • Requires a local Chrome/Chromium instance with CDP enabled, adding setup overhead
  • Scraping is primarily designed for Zhihu; extending to other platforms would need modification
  • Backtesting assumes a fixed risk‑free rate (2.8%) and default initial capital (1,000,000), with position rules that may affect comparability

README quick start

安装

Description

投资大V合订本:追踪投资大V历史预测,并用公开数据验证准确性

Related repositories

Similar projects matched by category, topics, and programming language.

Korben00
Korben00 GitHub avatar

no-gdid

no-gdid is a PowerShell-based tool to audit and block the Windows Global Device Identifier (GDID), a per-account device ID used by Microsoft and shared with law enforcement, based on verified reverse engineering.

PowerShell
241
LewdM3at
LewdM3at GitHub avatar

PC-Cartridge-System

A hobby project that turns 2.5" SATA SSDs into physical game cartridges for your Steam library, automatically launching games or custom actions when inserted.

PowerShell
219
gd03gd031
gd03gd031 GitHub avatar

Windows-GDID-Changer

A PowerShell script for Windows 10/11 that regenerates the Global Device Identifier (GDID) by removing local session tokens and forcing a new device registration with Microsoft servers.

PowerShell
149