Catalyst Surface Agent

Catalyst Surface Agent is a reusable autonomous trading engine for scheduled market events. Each week it scans a fixed universe of liquid stocks instead of waiting for a ticker or prompt. It reconstructs option surfaces, verifies event timing from independent sources, replays each candidate’s historical options, and promotes only setups that survive a frozen policy. Featherless runs a multi-model semantic committee that checks whether the catalyst remains intact and interprets the release. The models are deliberately bounded: they may veto risk, but cannot invent a trade, increase position size, relax a limit, or delay an exit. Alpaca MCP powers the entire lifecycle: account state, market clock, news, stock and option data, historical evidence, orders, positions, reconciliation, and portfolio results. In the measured deployment, the agent narrowed 64 stocks to one conditional AVGO earnings straddle, sized its exact maximum loss, entered autonomously, held through the release, and exited on schedule. AVGO moved 6.17% by the exit, validating the magnitude thesis, but the premium paid and post-earnings volatility collapse produced a loss. We publish that result without cherry-picking and use it to improve the next pricing policy. Every decision is written to a secret-redacted, tamper-evident hash chain. The live dashboard exposes the selection funnel, model committee, rejected strategies, broker lifecycle, measured outcome, and evidence behind the next iteration. Catalyst Surface Agent is not an AVGO bot: it can select a different qualified event next week or correctly remain flat.