A polymarket trading bot that detects temporary pricing inefficiencies across connected Polymarket sports markets.

This repository is not a commercial product.

It exists for one reason:

To share a real working trading system instead of publishing endless AI-generated articles that provide no practical value.

There are thousands of blog posts explaining arbitrage.

Very few people actually build one.

This project is my attempt to bridge that gap.

Everything inside this repository comes from building, testing, breaking, and rebuilding real trading infrastructure.

This bot will not magically print money.

I don't hide that.

Markets evolve.

Edges disappear.

Liquidity changes.

Competition increases.

This repository is a foundation, not a guaranteed income machine.

However, it contains many of the engineering ideas, mathematical models, and architecture that I used before becoming consistently profitable.

If you're interested in quantitative trading, prediction markets, market microstructure, or TypeScript architecture, I hope you'll learn something useful.

While building this project I also created:

STAKE-MATH

A Node.js library implementing Kelly-based stake sizing and probability mathematics.

Trading is not gambling.

Trading is probability management.

Without mathematics, risk management becomes guessing.

Many of the position sizing models used here are based on the Kelly Criterion, fractional Kelly, expected value, and bankroll optimization.

Those mathematical foundations have been responsible for far more of my long-term success than any individual trading strategy.

If you decide to experiment with this project:

• start with simulation mode
• understand every trade the bot makes
• test your own ideas
• use small position sizes
• never risk money you cannot afford to lose

Happy Trading ❤️

• ⚡ Real-time Polymarket CLOB streaming
• 🧠 Cross-line arbitrage detection
• 📈 Kelly Criterion stake sizing
• 💰 Paper trading & Live trading
• 🛡 Risk management engine
• 📊 Beautiful terminal dashboard
• 🧩 Modular architecture
• 📝 Structured logging
• 🚀 Production-oriented TypeScript codebase

Sports markets inside a single event should satisfy several no-arbitrage relationships.

Whenever one market reprices faster than another after new information arrives, temporary inefficiencies can appear.

The bot continuously:

Gamma API
      │
      ▼
Discover connected markets
      │
      ▼
Stream live CLOB orderbooks
      │
      ▼
Detect pricing violations
      │
      ▼
Risk evaluation
      │
      ▼
Kelly stake sizing
      │
      ▼
Submit limit orders
      │
      ▼
Track positions & PnL

Current arbitrage checks include:

• Complementary YES/NO pairs
• Totals ladder relationships
• Spread ladder relationships
• Moneyline vs Spread
• Three-way market sums
• BTTS relationships

By default the bot focuses on the markets where liquidity is strongest.

SPORT_FOCUS=nba,world_cup

Discovery logs look like:

Discovery refresh:

40 events
520 tokens

NBA: 10
World Cup: 30

Adding another sport only requires creating another profile inside

src/model/sportsRegistry.ts

git clone https://github.com/PolySports/polymarket-cross-line-arbitrage-bot.git

cd polymarket-cross-line-arbitrage-bot

cp .env.example .env

npm install

npm run start:sim

npm start -- --mode sim

npm start -- --mode live --confirm-live

npm start -- --event nba-lal-bos-2026-01-15

npm start -- --tag 100381

┌───────────────────────────────────────────────────────────────┐
│ Header                                                       │
│ Mode • WS Status • Balance • PnL • Uptime                    │
├───────────────────────────────────────────────────────────────┤
│ Tracked Markets            │ Opportunities                   │
├────────────────────────────┴──────────────────────────────────┤
│ PnL Chart                                           Exposure  │
├───────────────────────────────────────────────────────────────┤
│ Orders / Fills             │ Alerts                          │
└───────────────────────────────────────────────────────────────┘

[p] Pause

[f] Flatten

[q] Quit

Logs are written separately using Pino so they never corrupt the dashboard.

See .env.example.

Important variables:

The project sizes positions using the Kelly Criterion.

f* = (p − price)
     ───────────
      (1-price)

stake = bankroll
      × Kelly
      × Fractional Kelly

Position sizes are automatically clamped between:

• Minimum stake
• Maximum position
• Event exposure limits

Supported for:

• Locked arbitrage
• Relative value
• Hedged ladder trades

Default configuration uses Half Kelly (0.5) to reduce variance.

                Gamma REST
                     │
                     ▼
               Event Graph
                     │
                     ▼
              Market Classifier
                     │
                     ▼
           CLOB REST / WebSocket
                     │
                     ▼
             OrderBook Store
                     │
                     ▼
            Arbitrage Detector
                     │
                     ▼
              Risk Management
                     │
        ┌────────────┴────────────┐
        ▼                         ▼
 Sim Executor              Live Executor
        │                         │
        └────────────┬────────────┘
                     ▼
            Portfolio + Dashboard

Live mode requires:

PRIVATE_KEY

CLOB_API_KEY

CLOB_API_SECRET

CLOB_API_PASSPHRASE

and either

--confirm-live

or

CONFIRM_LIVE=true

Execution uses

@polymarket/clob-client-v2

against production endpoints.

src/

├── arb/
│   └── Arbitrage detection

├── config/
│   └── Zod configuration

├── core/
│   └── Engine

├── data/
│   ├── Gamma
│   ├── CLOB REST
│   └── WebSocket

├── exec/
│   ├── Simulation
│   ├── Live
│   └── Order Manager

├── model/
│   ├── Event Graph
│   └── Classifier

├── portfolio/
│   └── PnL

├── risk/
│   └── Exposure controls

├── ui/
│   └── blessed-contrib dashboard

└── util/
    ├── Logging
    ├── Math
    └── Rate limiting

npm test

• Additional sports
• Historical backtesting
• Strategy plugins
• Multi-exchange support
• Automatic parameter optimization
• Portfolio analytics
• Web dashboard
• Discord notifications

Contributions are welcome.

Whether you're interested in:

• quantitative trading
• prediction markets
• TypeScript
• market microstructure
• performance optimization

feel free to open an Issue or Pull Request.

This repository is provided for educational purposes only.

Nothing contained here should be interpreted as financial advice.

Prediction markets involve substantial financial risk.

Always test thoroughly using simulation before deploying capital.

Past performance does not guarantee future results.

If this repository helped you learn something new, consider giving it a ⭐.