13forge: AI Trading Safety Airlock

13forge separates AI ideation from financial execution. Generative models are phenomenal at discovering trading ideas and extracting market theses from news or prompts. However, they are not a safe place to put final authority over strikes, risk, or live broker payloads. In options trading, one confident hallucination or one missed risk calculation can liquidate an entire account. We built 13forge to give the AI a narrow job—to emit a constrained trading thesis—but it never gets a direct path to the broker. Before any request touches the Alpaca API, it must pass through a strict seven-gate Rust lattice. 13forge rebuilds the order from quoted market data and runs deterministic safety checks (Margin, Position-state, AI permission, Market stability, Trade structure, Maximum-loss). If a trade breaches a hard limit, it is refused by default, and the broker process is never started. The system proves what it refused, why it refused it, and exactly what did not happen next. The demonstration of a $2,525 maximum loss veto intercepting an oversized AI iron-condor is a massive step forward for autonomous trading safety. 🔗 Pitch Presentation: https://youtu.be/YO1xjnEtoF4 🔗 Technical Demo: https://youtu.be/IrLmPIXukyo 🔗 Live Portal: https://13forge-proof-portal.vercel.app/demo-portal/ 🔗 GitHub: https://github.com/deveraux-dev/AlpacaCOMP