Technology
A trading system has two jobs: act in the moment when prices move, and learn over time so tomorrow’s settings are better than today’s. GTS keeps those jobs separate so neither slows the other down. This page explains what each part does and how a new idea becomes live trading.
Decision models
Four model families run in production. Each one owns a single job—quoting, tuning, reading trends, or recognising bad conditions—so when something drifts, we know exactly which part to fix.
Quoting
Decides where to bid and offer on both sides of the market.
The system learns from past sessions how aggressively to quote and how much size to show. It is rewarded for staying near target inventory, so it hedges rather than letting positions run away.
Tuning
Adjusts settings while markets stay open.
Instead of shutting down for a manual retune, the platform tests small setting changes against recent results and only keeps improvements that still respect risk limits.
Trend reading
Finds structure in how prices and volume move together.
Patterns from one market inform another, so a move in index products can shape how the book responds in related crypto or ETF positions.
Conditions
Chooses what to trade together and when conditions turn hostile.
Thousands of pair relationships are ranked continuously. When volatility spikes, the system widens quotes and cuts size automatically before damage accumulates.
From idea to live trading
Nothing goes live because it looked good on paper. Each candidate is replayed on real market history, tested alongside production at zero size, and only then promoted—with a signed record of who approved it.
Every market tick and every fill we produce is stored so any day can be replayed exactly as it happened.
New ideas must be stated as a clear, measurable hypothesis. Ideas that cannot be tested do not move forward.
Candidates run against recorded sessions at real-world speed, including fees and queue behaviour they would actually face.
Survivors quote alongside production at zero size until their predicted behaviour matches what really happens.
Approved changes roll out to every connection without a maintenance window. Each change is signed and logged.
Infrastructure
We spend effort on the path between a market move and our response. Everything else is built to be easy to monitor and easy to change, so surprises stay in research—not on a live order.
Servers placed next to exchanges, one hop from the matching engine
The path from signal to order is kept simple and predictable. Surprises belong in research, not on the wire.
Research and live trading read the same history. Control updates never block the quotes that must go out now.
Models suggest improvements; risk limits have final say. Nothing reaches a venue without passing the same gate.
Nothing goes live without a replay run at production speed on recorded market data.
No setting change reaches a venue unless risk limits approve it, including changes suggested by the tuner itself.
No strategy is shaped around outside investor requirements, because there is no outside capital in the book.
Whether you are a venue, a counterparty, or just curious about the firm, reach out and we will point you to the right person.
