Crowd Excess

Crowd Excess Agent tests whether investor attention and price can move further than objective news evidence supports. It combines complete-day NAVER cross-border search attention with Alpaca prices, SPY-relative movement, volume, news, and option-chain data. OpenAI evaluates only the direction, materiality, and confidence of supplied headlines through a strict structured schema. Deterministic code owns scoring, option construction, sizing, account validation, and every risk gate. Seventeen real five-symbol sampled shadow scans ran across two US market dates on August 28 and 31, 2026. They produced 85 signal snapshots and 78 model assessments. No candidate passed every evidence, liquidity, and risk gate, so the account remained at $100,000 equity with zero P&L, drawdown, open risk, decisions, receipts, and positions. The public audit trail exposes source hashes, model response metadata, token usage, evidence, Crowd Excess residuals, failure-safe behavior, and the honest no-order outcome. Live trading, naked options, and browser-initiated execution do not exist.