About this project

btcbot is a read-only research bot for Kalshi's 15-minute Bitcoin prediction contracts. It polls public prices, estimates whether Bitcoin will finish higher than the current target using distance to target, time remaining, and recent movement, and flags a call when its estimate disagrees with the market price by at least 7 cents within the 70-90 cent band, confirmed over two minutes. It sends phone notifications but places no orders and requires no account or API key. The repository includes a live paper account with tracked balance, win rate, settled calls, and recent call history. The README reports 269 settled calls with an 84.0% win rate against an 80.0% break-even, while noting that roughly 100 settled calls are needed for meaningful statistics. It also documents a 63-day backtest of 329 trades at 88.4% accuracy against a 79.9% break-even, with caveats about limited historical data granularity. The project records rejected ideas and their measured results, including loss-avoidance learning, early exits, drawdown-based sizing, secondary price feeds, and chart indicators. Supporting files include a full learning report, research documentation, a literature review, raw Kalshi and Bitcoin data, and an order-book recorder. The bot runs on GitHub's infrastructure and can be executed locally with standard-library Python using check.py, with optional report and alert modes.