// architecture
How it works.
The system that decides, the guardrails that catch it, and the record we keep so you can audit the whole thing.
01
Agents debate every trade
Six adversarial agents score each signal on eight dimensions. If Red Team vetoes, no trade. No override. No "gut feeling."
02
Hard risk limits enforced in code
Max daily loss. Max position size. Correlation cap across the portfolio. These are not guidelines. They are constraints the bot cannot violate.
03
Every decision logged and journaled
Event-sourced journal. You can replay any day. If something broke, we can point to the exact tick and the exact agent output.
04
Walk-forward backtests, not curve-fits
Parameters get optimized on a rolling window and tested on unseen data. We publish the out-of-sample results, not the cherry-picked window.