Polymarket Pipeline V2

An AI-powered breaking news detector that classifies events against prediction markets and trades automatically when it finds edge.

Breaking News (Twitter / Telegram / RSS)
        ↓ (< 5 seconds)
Match to niche markets (< $500K volume)
        ↓
Claude Classification: bullish / bearish / neutral + materiality
        ↓
Edge detection + quarter-Kelly sizing
        ↓
Instant execution → SQLite log → calibration tracking

What Changed From V1

V1 scraped RSS feeds (5-60 min delay), asked Claude "what's the probability?" (wrong question for LLMs), and competed on high-volume markets (where every bot already operates).

V2 inverts all three: - Speed: Real-time Twitter/Telegram streams instead of stale RSS - Classification: Claude classifies "bullish or bearish?" instead of estimating probability — a task LLMs are actually good at - Niche markets: Only trades markets under $500K volume where the crowd is small and slow

---

Setup (2 minutes)

One-Command Setup

git clone https://github.com/brodyautomates/polymarket-pipeline.git
cd polymarket-pipeline
bash setup.sh

Manual Setup

git clone https://github.com/brodyautomates/polymarket-pipeline.git
cd polymarket-pipeline
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env

Add your keys to .env:

ANTHROPIC_API_KEY=sk-ant-...         # Required
TWITTER_BEARER_TOKEN=...             # Optional — real-time news stream
TELEGRAM_BOT_TOKEN=...               # Optional — channel monitoring
POLYMARKET_API_KEY=...               # Optional — live trading only

Verify

python cli.py verify

---

How to Use

V2: Event-Driven Pipeline (Recommended)

# Start the real-time pipeline — monitors news streams, classifies, trades
python cli.py watch

# Enable live trading
python cli.py watch --live

The watch command runs indefinitely. It connects to your configured news sources (Twitter, Telegram, RSS fallback), matches breaking headlines to niche Polymarket markets, classifies each with Claude, and executes trades when it finds edge.

V1: Synchronous Pipeline

# Single scan — scrape RSS, score markets, log signals
python cli.py run

python cli.py run --max 15 --hours 12

Live Dashboard

python cli.py dashboard

Backtest

# Validate the V2 strategy against resolved markets
python cli.py backtest

python cli.py backtest --limit 50 --category ai

All Commands

Command	What it does
python cli.py watch	V2: Real-time event-driven pipeline
python cli.py run	V1: Synchronous RSS-based pipeline
python cli.py dashboard	Live terminal dashboard
python cli.py backtest	Backtest against resolved markets
python cli.py calibrate	Classification accuracy report
python cli.py niche	Browse niche markets (volume-filtered)
python cli.py verify	Check all API keys and connections
python cli.py scrape	Test news scraper
python cli.py markets	Browse all active markets
python cli.py trades	View trade log
python cli.py stats	Performance + latency + calibration stats

---

Architecture

V2 Pipeline (Event-Driven)

news_stream.py      Real-time news — Twitter API v2, Telegram, RSS fallback
market_watcher.py   Polymarket WebSocket — live prices, niche filter, momentum
classifier.py       Claude classification — bullish/bearish/neutral + materiality
matcher.py          Routes breaking news to relevant markets
edge.py             Edge detection + Kelly sizing (V2: classification-based)
executor.py         Trade execution — dry-run + live CLOB orders (async)
pipeline.py         Event-driven orchestrator (asyncio)
calibrator.py       Tracks classification accuracy over time
backtest.py         Historical replay for strategy validation

Shared Infrastructure

logger.py           SQLite — trades, news events, calibration, latency tracking
config.py           All settings, API keys, thresholds
dashboard.py        Bloomberg Terminal-style live dashboard
cli.py              CLI — watch, run, backtest, calibrate, niche, verify, etc.

---

How It Actually Works

1. News Detection

Real-time streams from Twitter (filtered by keywords: OpenAI, Bitcoin, Fed rate, etc.), Telegram channels, and RSS fallback. Events are deduplicated and timestamped with receive latency.

2. Market Matching

Each headline is matched to active niche markets (<$500K volume) by keyword overlap. Only relevant markets proceed to classification.

3. Classification (The Key Shift)

Instead of "what's the probability?", Claude is asked: "Does this news make the market MORE likely to resolve YES, MORE likely to resolve NO, or is it NOT RELEVANT?"

This is a classification task — something LLMs are genuinely good at. Claude also rates materiality (0-1): how much should this move the price?

4. Edge Detection

If direction is bullish/bearish AND materiality exceeds threshold (default 0.6) AND the market price has room to move — that's a signal. Position sizing uses quarter-Kelly.

5. Execution

Dry-run by default. Live mode places orders via Polymarket CLOB API. Safety: $25 max bet, $100 daily limit.

6. Calibration

Every trade is tracked. As markets resolve, the system measures whether its classifications were correct. Accuracy by source and category informs future confidence.

---

Configuration

Setting	Default	What it does
DRY_RUN	true	Set to false for live trading
MAX_BET_USD	25	Maximum single bet
DAILY_LOSS_LIMIT_USD	100	Pipeline halts if breached
EDGE_THRESHOLD	0.10	Minimum edge to trigger trade
MAX_VOLUME_USD	500000	Only trade markets below this volume
MIN_VOLUME_USD	1000	Skip dead markets
MATERIALITY_THRESHOLD	0.6	Minimum materiality to act on
SPEED_TARGET_SECONDS	5	Target news-to-trade latency

---

Safety

• Dry-run mode ON by default
• $25 max single bet, $100 daily limit
• Quarter-Kelly position sizing
• Niche market filter prevents competing against sophisticated bots
• Calibration tracking — auto-detects if strategy accuracy drops
• All API keys in .env, never committed

---

Built by @brodyautomates

---

Disclaimer

This project is for entertainment and educational purposes only. It is not financial advice. The authors are not responsible for any financial losses incurred through the use of this software. Prediction market trading carries significant risk — you can lose money. Never trade with funds you cannot afford to lose. Past performance of any strategy does not guarantee future results. Use at your own risk.