AutoPredict

AutoPredict is a framework for prediction market trading agents. It connects to live Polymarket data, lets you supply your own probability estimates, and evaluates trade opportunities with execution-aware metrics.

Quick start

git clone https://github.com/howdymary/autopredict.git
cd autopredict
python -m pip install -e .

# Scan live markets
python predict.py

# Find structural edges in multi-outcome events
python predict.py --events

# Test your own prediction on a specific market
python predict.py --fair 0.60 <condition_id>

See QUICKSTART.md for a full walkthrough.

What it does

• Live market scanning: Fetches active markets and real order books from Polymarket (no auth needed for reads)
• Event-level analysis: Finds multi-outcome events where sibling prices don't sum to 1.0 (structural mispricing)
• Execution-aware agent: Given your fair_prob, evaluates edge, spread, liquidity, and book depth before recommending a trade
• Configurable strategy: All agent parameters are JSON-tunable (edge thresholds, risk limits, sizing)
• Backtesting engine: Test strategy changes against market data with slippage and fill rate simulation

Core pieces

• predict.py: Live market scanner and agent runner — the main entry point
• agent.py: The mutable trading agent
• market_env.py: Order book simulation and execution metrics
• autopredict/markets/polymarket.py: Polymarket API adapter (Gamma + CLOB)
• strategy_configs: Tunable strategy parameters
• run_experiment.py: Backtest loop for offline evaluation

What it measures

Three groups of metrics:

• Epistemic: brier_score, calibration_by_bucket
• Financial: total_pnl, sharpe, max_drawdown, win_rate
• Execution: avg_slippage_bps, fill_rate, market_impact_bps, implementation_shortfall_bps

Documentation

• QUICKSTART.md
• docs/ARCHITECTURE.md
• docs/STRATEGIES.md
• docs/BACKTESTING.md
• docs/METRICS.md
• docs/fair_prob_guidelines.md
• docs/DEPLOYMENT.md
• docs/TROUBLESHOOTING.md

Design philosophy

The agent does NOT generate predictions. It optimizes execution given your forecast. If you think a market is at 54% and should be at 60%, the agent decides whether to trade, how much, and what order type — based on spread, depth, and your risk limits.

The forecasting is your job. The execution is the agent's job.

License

MIT. See LICENSE.