
Catalyst Router is built to compete on paper-trading P&L without treating an LLM as an ungoverned trader. It screens a daily universe of liquid equities with Alpaca IEX data, scores completed 15-minute bars using a checksummed XGBoost model, and converts only high-confidence long or short signals into trade proposals. Every proposal must pass fresh-quote, spread, buying-power, exposure, drawdown, and position-limit checks before the deterministic Risk Governor can submit an idempotent Alpaca bracket order with broker-hosted exits. The project demonstrates deep Alpaca integration across Paper Trading, Market Data, News, alpaca-py, and a judge-friendly CLI for safe paper-account inspection and dry-run demonstrations. Its differentiator is a clear separation of intelligence from capital authority: Bedrock extracts structured news events in shadow mode, the model supplies direction and confidence, and deterministic code retains all sizing and execution authority. A delayed, sanitized public flight recorder makes live agent behavior auditable, showing decisions, abstentions, vetoes, route selection, risk usage, drawdown, and P&L against cash. This makes Catalyst Router a credible, transparent, and operationally robust autonomous trading system rather than a backtest-only demo.
4 Sep 2026