Zurich · Cape Town · Remote
Oresund
Matching engines, derivatives systems, and market infrastructure. Mission-critical software where latency, determinism, and correctness under failure all have to hold at once.
01 — Domain
Markets are a substrate. These are the primitives we think about every day: what they are, why they are hard, and where they tend to break when the real world gets noisy.
The heart of an exchange. A deterministic state machine that pairs bids and offers under strict price-time priority. Every microsecond of tail latency is a story someone has to answer for.
Options, futures, swaps, exotics. Pricing is only half the job. The rest is risk: Greeks, stress scenarios, portfolio margin, wrong-way exposure. The math is famous; the edge cases are not.
Markets for contingent claims. The mechanism design problem is liquidity under thin flow; the engineering problem is fair resolution under ambiguous reality. Oracles, AMMs, order books, and everything in between.
Block execution without printing intent to the tape. Crossing networks, conditional orders, midpoint pegs. Built to minimize market impact while staying inside the regulatory envelope.
How prices actually form. Maker-taker flow, queue position, adverse selection, toxicity. The invisible physics underneath every quote screen.
Post-trade is where risk gets netted, margin gets called, and assets actually change hands. CCPs, DvP, atomic settlement on chain. Invisible when it works, catastrophic when it does not.
02 — Primer
Not services. Not offerings. Just the concepts that decide whether a market system is trustworthy once it leaves the happy path.
Determinism
A matching engine is a promise: identical inputs produce identical outputs. That is what lets participants trust the venue and what lets operators reason about failure without rewriting history after the fact.
Non-determinism is not a cosmetic bug. It is a regulatory, financial, and reputational liability wearing the costume of a race condition.
Tail Latency
Mission-critical systems are judged at p99.99 and beyond, because the one slow event is exactly the one the counterparty remembers. Garbage collection, kernel scheduling, cache misses, NUMA boundaries: all of them live in the tail.
Low latency is not one number on a slide. It is an argument about worst-case behavior, resource isolation, and whether the system stays composed when the order flow is not polite.
Failure Paths
The interesting moments are the awkward ones: partial fills in a crossed book, reconciliation after a link drop, forced liquidation in a gap move. Systems that handle the boring path well and the rare path badly are how losses become public.
Correctness under failure is what keeps post-trade, risk, and customer trust aligned after the first assumption has already failed.
03 — Contact
We engage with a small number of firms at a time. If you are building at the edge of a market, pre-launch, scaling, or rebuilding something critical, let us know.