Performance is measured across three independent engines operating on a paper Alpaca account. All trades are recorded at the moment of order submission; all outcomes are recorded at the moment of position closure. No survivorship bias is applied — every trade entered appears in this record regardless of outcome.

Signals are sourced from federal disclosure data (congressional filings via STOCK Act, SEC Form 4 insider disclosures, 13F institutional filings, government contract awards, and lobbying registrations) combined with observed market structure (options flow and dark pool print data). Each signal is scored 0–100 for conviction at the time of evaluation, based on source credibility, recency, cross-source confirmation, and contextual market factors.

Position sizing follows a disciplined fractional Kelly approach. No single position exceeds the configured maximum allocation. A drawdown circuit pauses new entries if portfolio value falls 8% from its peak. Daily loss limits provide an additional control layer.

Win rate is calculated as the proportion of closed trades with positive realised P&L. The 95% confidence interval is computed using the Wilson score method. Expectancy is the probability-weighted average outcome per trade (win rate × average win + loss rate × average loss).

Historical performance is preserved in a Postgres database with immutable closed-trade records. The equity curve represents cumulative realised P&L by trade-close date. All figures are paper-trading results — no real capital is deployed.