5 of 10 limits could fire in the last cycle (2026-11-06 13:47 UTC, 11 symbols); the others are off or lack the input they read. The limits are the file below, not code: the paper cycle reads it
each run, an edit takes effect only once sealed, and a file that does not load stops the cycle
deciding anything rather than falling back to a guess. Policy digest
68ceeadb52fc4ce0.
| Limit | Value | Last cycle |
|---|---|---|
| max position notional largest notional one order may carry | $50,000 | can fire |
| max unhedged notional largest exposure carried while no hedge is available | $20,000 | can fire |
| min confidence to trade stated confidence below which no order is sent | 0.55 | can fire |
| max gross exposure notional total gross notional across the book, order included | $150,000 | can fire |
| max signed exposure notional net long-minus-short notional across the book | $100,000 | can fire |
| max margin usage ratio venue margin ratio cap; 1.0 is the liquidation line | 0.5 | cannot fire |
| factor exposure limits cap on the book's exposure to each named factor | off | cannot fire |
| max scenario loss pct floor on the book's move in the worst benchmark shock | off | cannot fire |
| max liquidation cost bps ceiling on the worst position's forced-exit slippage | off | cannot fire |
| assumed payoff win/loss ratio used by Kelly sizing once calibration passes | 1 | parameter |
| session horizon bars assumed holding period for the volatility throttle | 24 hours | parameter |
| min symbol realized pnl floor on one symbol's realised PnL over the window | off | cannot fire |
| symbol underperformance window minutes window for the per-symbol PnL floor | 1440 minutes | parameter |
Read from
data/risk_policy.json and
data/risk_policy_live.json;
enforced by
risk/constitution.py.