CUATS ORCA

ALERT state
BORDERLINE state
OK state
HALT state
Simulated exchange with live agent trading
Visual session summary
Session summary (snippet)

[markdown]

Inspiration

An outstanding technical need for quant firms is the establishment of a robust means to enforce pre-trade checks, monitor data quality and drift, track lineage and unauthorized changes, and trigger safe-mode or kill-switch action.

What it does

The CUATS ORCA (Orderbook Risk and Control Analyser) is a system that monitors the integrity, model-input drift, order-path health, and PnL, then automatically downgrades or halts trading, issuing corresponding audiovisual prompts, before a bad feed, bad deployment, or regime shift becomes a loss or a compliance event.

How we built it

The system has three main layers:

Exchange / matching engine: a limit order book supporting limit, market, IOC, and FOK orders with price-time priority.
Agent-based market simulation: a set of trading agents that generate realistic and adversarial order flow.
Risk control plane: a monitoring layer that classifies the market into operational states including OK, ALERT, BORDERLINE, and HALT.

Challenges we ran into

Keeping the simulation realistic enough that the monitor was meaningful, not just noisy.
Calibrating thresholds so the system did not stay permanently in an alert state.
Syncing the Python live dashboard and browser-based server so both used the same BTC + GBM setup.
Debugging live UI issues while keeping the exchange, agents, and control logic consistent.

Accomplishments that we're proud of

Building a full end-to-end stack: exchange, agents, risk-control logic, and live interfaces.
Creating a system where market microstructure actually drives interpretable control states.
Making the output understandable in real time through clear audio/visual signalling.
Turning a technical risk-monitoring idea into something demo-able and intuitive.

What we learned

Good monitoring is not just about detection, but about calibration and interpretability.
Small assumptions in the simulator can have a big effect on downstream thresholds and UI behaviour.
Agent-based markets are a strong way to test control logic under both normal and adversarial conditions.
Live systems are much easier to trust when they can explain why a state changed.

What's next for QuantiHack 2026

Connect ORCA to real or streamed market data.
Add stronger deployment lineage and change-tracking around models and trading logic.
Improve incident explainability and historical replay.
Extend the control plane from simulation alerts toward real gateway actions and automated safeguards.

[\markdown]