Veritas-Agent

VERITAS is an execution-aware autonomous AI trading agent built for the Alpaca AI Trading Agents Hackathon. Its goal is not simply to predict market direction, but to determine whether a proposed options trade is trustworthy enough to execute. VERITAS focuses on defined-risk short-DTE credit spreads on SPY and QQQ. A deterministic candidate engine first generates valid options spreads using predefined constraints. An AI reasoning layer then evaluates and ranks those opportunities using market context, technical signals, volatility information, and the strategy's rules. The AI does not have unrestricted control over execution: every proposal must pass deterministic mathematical validation and a multi-layer risk gate before an order can be submitted. The project's core principle is: “LLM proposes, math disposes.” The system independently recalculates option spread width, credit, maximum loss, breakeven, position size, and other critical values instead of trusting the language model's numerical output. It also applies an Execution Reality Layer that evaluates whether a trade remains attractive after conservative assumptions about liquidity, slippage, and paper-trading execution behavior. A key feature of VERITAS is its Execution Confidence Score, which measures the quality and trustworthiness of each candidate based on factors such as quote quality, liquidity, data freshness, spread characteristics, and execution assumptions. Low-confidence opportunities can be rejected or traded at reduced size. The agent also continuously reconciles positions and account activity, polls for options activities that are not delivered through streaming events, and uses a hierarchical risk system with per-trade limits, daily loss protection, position-count limits, portfolio heat, liquidity checks, DTE constraints, and an emergency kill switch. The default strategy avoids holding options into expiration and prioritizes controlled, defined-risk exposure.