BaDing: Barokah Trading

BaDing is an evidence-gated 5-minute intraday trading agent designed to solve the problem of unfalsifiable confidence in AI trading demos. While most systems execute trades even when evidence is weak, BaDing is built to refuse bad trades and transparently journal the exact mechanical reason for the denial. The architecture operates on five strict rules: AI proposes: The language model reasons and forms a thesis, but it cannot size a position or execute a decision. Quant provides evidence: It evaluates market data deterministically using a continuous, fixed 277-bar causal window. Deterministic code decides risk: A rigid firewall enforces mechanical rules and actively denies any trade lacking statistical validation. Broker truth decides state: Local memory is never trusted over broker reality. The journal proves what happened: Every decision cycle is immutably logged before the AI is even called. During pre-registered testing, BaDing's candidate strategy yielded a negative expectancy of −0.128R out-of-sample. Instead of parameter mining to artificially force a profitable backtest, the system logged the failure. The deterministic risk layer now mechanically denies execution authority for this strategy, rejecting entries with a CANDIDATE_NOT_VALIDATED reason code in both shadow and paper modes. BaDing demonstrates that an autonomous agent's highest value is not in predicting the market, but in remaining trustworthy and safe when it lacks a statistical edge.