How we work

Nothing is traded on a hunch. Each strategy is written down as a model with the data it reads, the price it computes and the conditions under which it acts, and every order it places is sent by software.

That makes every decision reproducible. When a trade goes wrong, we can see what the model saw at that moment and why it acted.

Before a strategy trades, it is scored on recorded markets the way it would actually have traded them: using only what was known at each moment, walked forward day by day, at the prices and sizes that were really available, after fees.

Results are judged on whole days rather than single trades, because the events of one day tend to move together. When live results part from what the model expected, the model is what gets re-examined.

We record the markets we trade and the sources that move them, every update stamped on our own synchronised clocks, so any moment can be replayed exactly as it was seen.

The systems that receive data and send orders are our own, running on machines placed for the markets they serve. Every leg from an event to our order is measured, and the slow ones are where the engineering goes.

An event contract pays on what its settlement source reports, so the source is the most valuable input there is. Wherever we can, we take each event from the league, the game publisher or the venue itself rather than from a copy further down the line.

Data we take in is used for our own trading only. It is not displayed, published or passed on.

We trade only our own capital. Each strategy runs within limits on its size and on what it may lose before it stops, and a strategy is given more only once it has earned at the size it runs.