For weeks almost nothing happens. Then the event gets close and thousands of people arrive together, in the last few minutes, all doing the same thing at the same instant.
A trading system is judged there. Not on how it holds up on average, but on what it does in the minute everything counts.
Where it gets complicated
The order
Operations have to be processed in the exact sequence they happened, with no exceptions. It isn’t a design preference: it’s the condition for the market to be correct. A system that orders things wrong isn’t slow, it’s wrong.
The speed
Between the operation and the confirmation there’s a margin users perceive, and on which they build trust in the platform. The techniques that suffice for a normal application don’t get you there.
The match
Market orders, limit orders, marketable-limit orders: each has its own matching rules and its own edge cases that only show up once a lot of people are in at once. It’s the part that always gets underestimated, because described in words it sounds simple.
The exposure
How much each person can still commit depends on positions that change at the very moment the calculation is made. And the number can’t be rounded down, or up.
The rules
Fees tied to incentives and to market state, calculated during the operation and not after. Changing the formula means being able to reconstruct what would have happened under the old one.
And underneath it all: every single operation has to be traced so that, months later, exactly what happened and why can be reconstructed. Not out of caution — because sooner or later someone asks.
Before
A market where people take positions on future events, aimed at a professional audience. No specialized vendors to turn to: the few who've built one built it for themselves.
After
The order-matching engine, the exposure and fee logic, with full tracking of every operation. Project ongoing.
Why it matters
For a back-office system or an e-commerce site, decades of solutions already exist: whoever starts out has an example to follow somewhere. Here public references are few, recent, and almost always incomplete on the parts that matter.
It means the hard decisions have to be made, not looked up. And whoever has already made them once starts from a different point.
The difference isn’t how fast you write code.
It’s knowing which questions to ask first.
If the platform doesn’t exist yet, the first useful thing isn’t a quote: it’s understanding what the hard decisions are and what they cost — Build Plan. If it exists and needs to hold up until the first real event arrives, the door is Scale.
If instead you already know what needs doing and want to see how we work before committing, there’s the Dry Run: two weeks, and it ends there.