Technology

The fastest way from a fact in the world to a fair price.

Event markets move on information that is not on any exchange feed: a press release, a weather model run, an injury report, a vote count. Our platform is built around that fact.

Event ingestion

Thousands of primary sources (statistical agencies, central banks, election authorities, meteorological services, official league feeds), normalised into a single timestamped event stream with provenance on every field.

≈ 4.2bn events / mo

Fair-value engine

Category models and market-implied signals combined into a single probability per contract, with uncertainty bands that drive spread and size.

Re-prices in < 1 ms

Quoting & execution

Venue adapters for regulated exchanges and on-chain order books, with inventory skew, self-trade prevention and kill switches enforced in the hot path.

38 ms median news-to-quote

Resolution library

A structured database of contract rules, sources and edge cases, mapping equivalent questions across venues and flagging basis risk before it becomes P&L.

60k+ contracts mapped

Research platform

Point-in-time data, a backtester that replays order books and news together, and a scoring service that grades every model on every resolved contract.

Every model, every outcome

Latency, where it matters

Scheduled news is a race. Unscheduled news is a judgement.

For scheduled releases (inflation prints, rate decisions, official results) we parse the primary source directly and update quotes before most participants have loaded the page. For unscheduled news, speed matters less than being correctly sceptical, and our models are tuned to tell the difference.

Fig. 4. Distribution of news-to-quote latency across scheduled releases. Most mass sits in the first few tens of milliseconds; the tail is human-reviewed events.

Engineering culture

Boring where it should be. Ambitious where it pays.

The hot path is written in Rust and C++, the research stack in Python, and everything is reproducible from a single point-in-time data layer. We run on our own hardware close to the venues that reward it and on major public clouds (AWS, GCP) for research compute.

We care about correctness more than cleverness: typed event schemas, property-based tests on pricing code, replayable incidents and a strong preference for systems a new engineer can understand in a week.

Engineering roles are open.

Layer	Stack
Hot path	Rust · C++20 · kernel-bypass networking
On-chain	Rust · EVM tooling · custom indexers
Research	Python · Polars · JAX · PyTorch
Data	Point-in-time lake · columnar storage
Infra	Colocated hardware · AWS · GCP