hardDesignScenarioSeniorStaffLead
How would you put mandatory pre-trade risk checks on the order path when the desk will not accept the latency?
Pre-trade risk checks must stay on the order path without adding unacceptable latency. Keep static checks in-process, maintain bounded counters for local exposure and grant budgets for global limits, with a kill switch that fails safe. It also connects market access to the point an interviewer is testing.
7 min readUpdated 2026-08-15
Answer
Short answer
Put pre-trade risk checks on the order path by making them cheap, not optional. Static checks should be in memory, account-level counters should update atomically, and global limits should grant bounded budgets to gateways. Anything that cannot be proven safe should fail closed, and the kill switch must cancel or block orders even when other components are degraded.
Was this answer helpful?
Likely follow-ups
- Which checks would you be willing to evaluate on data that is a few hundred milliseconds stale, and how do you justify that to a regulator?
- How do you prevent two gateways from each approving an order that together breaches a single account limit?
- What is your fail-safe posture if the limit-state store is unreachable at start of day versus mid-session?
- How would you test the kill switch in production without stopping the desk?
Related questions
- An upstream feed resets ten thousand of your prices to a penny and orders start arriving. What should have stopped it, and what do you do with the orders that got through?hardAlso on kill-switch6 min
- A configuration change that passed review takes the entire fleet down in under a minute. What in your design should have limited the damage?hardAlso on kill-switch7 min
- Design a distributed lock manager for a high-frequency trading platform where microsecond latency is critical and stale locks can result in massive financial losses.hardAlso on low-latency3 min
- How do you design a global skill-based matchmaking system that prevents grandmasters from stomping novices without making them wait in a 20-minute queue?hardAlso on low-latency3 min