The framing has become familiar. Digital asset exchanges, prediction markets and retail-driven platforms have normalised continuous trading. Traditional venues, with their nightly batch cycles and weekly maintenance windows, are now playing catch-up as they extend hours, tokenise assets and reach for new distribution models. The conventional answer is to point at the matching engine and ask whether it can run around the clock.
That, several panellists at a recent TradingTech Insight webinar argued, is the wrong place to be looking.
The session, “From 24/7 to Event-Driven: Engineering the Next-Generation Exchange Platform,” sponsored by Adaptive and moderated by Mike O’Hara, Editor at A-Team Group, brought together Chris Walsh, Head of Trading Systems at EDX Markets; Dominik Eberle, Principal Product Owner at BX Digital; and David Smith, Global Head of Systems Engineering at Adaptive. Across nearly an hour of discussion, a clear thesis emerged: the matching engine is, by industry standards, a relatively well-understood problem. The harder questions sit in the surrounding infrastructure – surveillance, clearing, member connectivity, corporate actions, settlement workflows and post-trade reconciliation – and that is where most operators have not yet built for continuous operation.
The matching engine is the easy part
Smith made the point directly. Most matching engines have always been implemented as event-driven systems and are reasonable candidates for migration to continuous operation. The complications, he explained, lie elsewhere, in the infrastructural concepts and business-related tasks tied to calendar days. Risk processing, pre-trade checking and the broader post-trade workflow all carry assumptions about end-of-day boundaries that don’t survive contact with continuous operation. Couple those assumptions with legacy implementations maintained over very long periods, and operators looking to extend trading hours often find themselves drawn into wholesale platform modernisation rather than incremental change.
Eberle, whose role at BX Digital – Switzerland’s first FINMA-regulated DLT trading facility – gives him a clear view of the post-trade layer, brought the issue to life with an example that went well beyond the exchange’s own systems. In Switzerland, instant payments have created a situation where a customer can pay on a Saturday, see the payment confirmed instantly, and yet find that the value date sits two days later. “Did I really pay? Yes, no. What is this?” he asked. The point lands precisely because payments aren’t even on the exchange’s own infrastructure – they sit on someone else’s calendar – but they break the customer experience anyway. Multiply that across corporate actions, settlement cycles, member onboarding and reporting, and the scale of the surrounding-systems problem becomes clear.
Walsh, who has now built two centrally cleared crypto exchanges around 24/7 operation – first at ErisX and now at EDX – framed the architectural requirement in terms that reach beyond uptime. The system, he argued, has to behave as continuously on, continuously correct, not just at the matching engine but across all the services around it. Failover semantics matter as much as availability: when there is no maintenance window to hit a reset button, recovery has to be seamless, with no ambiguity in the state of the system that has just taken over. That principle of continual correctness, Walsh suggested, is what determines whether the rest of the platform – risk, surveillance, recovery – can be trusted at any given moment.
Cross-pollination, in both directions
Both Smith and Walsh credited the digital asset and prediction market world with adopting modern consensus algorithm implementations and replicated state machine architectures earlier than most traditional venues. Those choices, Smith observed, deliver the mixture of reliability, fault tolerance and performance that becomes more or less mandatory if a venue is running around the clock. Cloud-native deployment is a related strength, though Smith was clear that retrofitting cloud architectures onto trading workloads is harder than it looks.
The traffic also runs the other way. Walsh’s framing of EDX as bringing traditional market structure and discipline into a 24/7 trading environment captures the second direction of travel: institutional crypto venues importing the rigour around resilience, surveillance and governance that the regulated markets have built up over decades. The discussion explored both flows in some depth, including where the lessons transfer cleanly and where they don’t.
What the audience said
Two audience polls run during the session reinforced the picture. Asked where their organisation sat on the exchange technology refresh spectrum, the largest group – by some margin – said they were planning a refresh within the next 12 to 24 months. No respondent had recently completed a major platform overhaul, which Smith took as evidence that the current wave of architectural reassessment is still in motion rather than behind the industry.
A second poll, on the single biggest challenge in modernising exchange technology, was led by maintaining resilience and stability during the transition, with performance close behind. That finding prompted some of the more interesting exchanges of the session , including Walsh’s reflection on his two successive matching engine migrations at EDX, and Smith’s observation about why time-to-market for new products didn’t score higher in a year when prediction markets are launching contracts around individual events.
The architectural question the panel converged on
Asked what single architectural decision a newly licensed venue operator must get right to launch a 24/7 multi-asset platform within twelve months, the panel converged with unusual unanimity on a single foundational choice. One that, in Walsh’s framing, determines how resilient the venue is, how cleanly it recovers from failure, and how much operational overhead the operator carries for the life of the system. Get it right, and everything else gets simpler, not just the matching engine, but the surrounding services too. Get it wrong, and the operator ends up compensating for it everywhere else.
The specifics of that decision, the engineering culture required to operate it, and the trade-offs between buying foundational infrastructure and building it in-house formed the spine of a discussion that also covered vendor lock-in, the hardest parts of migrating off a monolithic platform without downtime, and whether regulators are keeping pace with the architectural shifts now under way.
What the conversation made clear is that the headline framing – moving from 24/7 to event-driven – only captures part of the architectural shift now under way. The matching engine, for all the attention it attracts, is not the binding constraint. The real work is in everything around it.
The full discussion, including Walsh, Eberle and Smith’s specific recommendations on the day-one architectural decision, the buy-build boundary and the practical mechanics of migrating legacy platforms without a big-bang cutover, is available on demand via the TradingTech Insight webinar archive.
Subscribe to our newsletter
