Every step between market data arriving and an order leaving has a cost: parsing, decision, serialisation, the wire. We treat the total like a budget and account for every line item, the way a careful CFO treats spend. Nothing gets to be slow anonymously; if a step takes time, it has to say so, out loud, in the profile.
The surprise, for most engineers who join this world, is how often the answer is not exotic hardware but subtraction: removing a queue, a copy, a lock, a layer that was only ever there by habit. Fast systems are mostly systems with less in them. The glamorous stuff, kernel bypass, FPGAs, exotic NICs, comes after the boring stuff, and the boring stuff is where most of the budget hides.
Why a budget beats a race
A race has one winner and burns everyone else's money. A budget is something every team can run: know where the time goes, spend it deliberately, and refuse to pay for anything twice. It turns "be fast" from a slogan into a series of small, checkable decisions, each one measured before and after. That's a culture, not a purchase.
The habit transfers to any engineering anywhere, which is why we interview for it even in candidates who've never touched a market. If you can tell us where the time went in the last system you built, we'll teach you the market part.
Sound like your kind of accounting? We're hiring engineers.