voltage-kalshi
Live volatility trading infrastructure for Kalshi BTC markets.
A real-capital trading loop that combines live market ingestion, short-horizon feature engineering, model-driven signals, and operator-aware execution.
Kraken WebSocket ingestion feeds a feature pipeline for short-horizon BTC movement.
Model signals combine tree-based structure with sequence-oriented forecasting ideas.
Execution logic sits behind the signal layer so risk, confidence, and market state can gate action.
Context
Kalshi BTC contracts are not the same as spot BTC trading. The useful question is often whether volatility and market structure imply a mispriced binary outcome.
The project is designed around live uncertainty: noisy streams, changing order books, execution timing, and the need to understand when not to trade.
Constraints
Live capital makes false confidence expensive.
Backtests are useful but insufficient because fills, latency, and market regime shifts change the real behavior.
The system needs enough observability to explain why it enters, exits, or stays flat.
Tradeoffs
Kept the first version focused on a narrow market instead of generalizing across every Kalshi contract.
Preferred interpretable operational signals over a black-box model-only story.
Accepted a smaller live scope to keep risk bounded while the data and execution loop matured.
Outcomes
Produced a coherent real-capital trading system rather than a notebook-only model.
Created a clear proof point for market-data ingestion, feature engineering, and execution-aware AI systems.
Lessons
Trading systems need operator surfaces as much as predictive models.
The strongest edge often comes from respecting constraints earlier than the model architecture discussion.