Execution & risk

Risk engine

Deterministic risk limits outside the agent layer, position sizing bounded by executable liquidity, and the global risk states that gate execution.

Risk rules exist outside the agent layer. The Risk agent may interpret conditions, but deterministic code enforces the limits — and it can block execution.

Risk limits

Limit	Bounds
Per-trade value	Maximum notional per trade
Capital allocation	Maximum percentage of available capital
Daily loss	Maximum loss in a day
Weekly drawdown	Maximum peak-to-trough loss in a week
Concurrent positions	Maximum open positions
Exposure per chain / exchange	Caps spread across venues
Minimum liquidity	Required liquidity to enter
Maximum slippage / spread	Execution-quality bounds
Holder concentration	Maximum top-wallet concentration
Contract-risk score	Maximum tolerated contract risk
API error rate	Trips on degraded connectivity

Position sizing by executable liquidity

Size depends on executable liquidity, not reported market capitalization — a large market cap with a thin pool still caps the position. The controls are deterministic and per-strategy:

maximum_position_usd
maximum_position_as_percent_of_pool
maximum_expected_price_impact_percent
maximum_entry_slippage_percent
maximum_exit_slippage_percent
minimum_exit_liquidity_usd
minimum_unique_traders
maximum_top_holder_percent
maximum_creator_holder_percent

position_sizing_v1

Risk-first quantity

Q u an t i t y = \frac{E q u i t y \times R i s k P er ce n t}{∣ E n t r y - S t o p ∣ + S l i pp a g e _{u ni t} + E x i tF e e _{u ni t}}

Start from the loss you accept, then the stop distance (widened for slippage and fees) fixes the units. The full derivation and the min(…) of every limit is on Position sizing & stops.

risk_metrics_v1

Portfolio heat

P or t f o l i oH e a t = \frac{\sum _{i} P l ann e d L os s _{i}}{E q u i t y}

Total open risk as a fraction of equity — the system-wide cap that throttles new entries. See Performance & risk metrics.

Risk states

A single global state gates execution across the whole system, escalating as risk rises:

NORMAL
Full execution within limits
CAUTION
Tightened sizing and extra checks
RESTRICTED
Reductions and approved exits only
HALTED
No new orders; open positions managed
EMERGENCY
Kill switch — cancel/flatten, human reactivation required

Circuit breakers escalate the state automatically on triggers such as the daily/weekly loss limit, max consecutive losses, stale data, abnormal spread or slippage, failed reconciliation, clock drift or repeated order rejection. No model may override a HALTED state.

Deterministic, not advisory

The Risk agent surfaces context and arguments, but the limits above are hard-coded and versioned (risk_policies). A breach raises a risk_event with an automatic action — block, reduce, exit or halt — independent of any model’s judgment.