Autonomous AI Agents & Rule Bots: Deterministic Risk Shield & Paper-to-Live Checklist

When you configure an AI Agent or Rule Bot on Evident, no LLM proposal is ever routed directly to an exchange. Every decision first passes through a Deterministic Risk Shield.

1. What the Risk Shield Enforces

1. Mandatory ATR-Bounded Stop-Loss: If the model omits a stop, one is placed at 2.0× ATR. Stops tighter than 0.5× ATR are widened to prevent noise sweeps; stops wider than 6.0× ATR are clamped.
2. **Risk-Based Position Sizing (l_risk): Regardless of stop distance, notional size is scaled so a stop-out loses at most your configured risk percentage (e.g., 1.0%** of equity):

$$\text{Notional Size (\$)} = \frac{\text{Equity} \times \text{Risk \%}}{|\text{Entry} - \text{Stop}| / \text{Entry}}$$

1. Liquidation Guard: If proposed leverage places the liquidation price inside the stop-loss distance, leverage is automatically stepped down.
2. **Daily Loss Circuit Breaker (l_daily):** Once realized intraday loss hits your daily cap (e.g., 3%), new entries are blocked for the remainder of the UTC day.

2. 14-Day Paper Audit Before Live Routing

• Review at least 20 decision cycles in the Decision Journal.
• Check how frequently the risk engine logs size_cut_by_risk or low_confidence.
• Only connect trade-only API keys (withdrawals disabled) or an isolated Hyperliquid agent wallet.