Australia version. Adds the Australian requirements to raise in banking, insurance and super interviews: APRA CPS 230 tolerance levels, the 24 and 72-hour notices, OAIC breach rules and local regions. US version

System Design Interview Framework (Australia)

A timed system design framework for Australian tech interviews, with APRA CPS 230 tolerance levels, notifiable data breaches and Sydney to Melbourne replication.

The interactive tool loads here when JavaScript runs.

About this tool

What is System Design Interview?

The four steps on this page (requirements, high-level design, detailed design, scaling) suit a 45 to 60 minute round in Sydney or Melbourne as much as anywhere. What changes at an Australian bank, insurer or super fund is APRA's prudential standard CPS 230 on operational risk, which commenced on 1 July 2025. It makes regulated entities keep a register of critical operations and set three tolerance levels for each: the longest disruption they would accept, the most data they would lose, and the minimum service they would keep running on alternative arrangements.

Those three tolerances are a ready-made set of non-functional requirements. The first is a recovery time, the second a recovery point, and the third a degraded mode your design has to describe. CPS 230 lists payments, deposit-taking, custody, settlement and clearing as critical for banks, and claims processing for insurers, so a prompt about any of those is a prompt about tolerances. Personal data brings in the Notifiable Data Breaches scheme run by the OAIC.

The worked examples below put numbers on both. Ticks on the framework and topics are saved in your browser only.

  • Four-step framework with time boxes, steps and tips
  • Six practice prompts with key components and focus areas
  • About 40 core topics with short explanations, including CAP, QUIC, gossip and Merkle trees
  • Checklists for framework steps and core concepts, saved in your browser
  • Link to an Excalidraw-based diagram tool for architecture sketches
Read the guide: System Design Interviews: A Four-Step Method with Real Numbers

Updated

How to use System Design Interview

  1. Open the Framework tab and read the four steps with their time boxes before your first practice session.
  2. Pick a prompt from the Practice Questions tab and set a 45 minute timer.
  3. Work through the steps in order: requirements and estimates, then a high-level diagram, then one or two components in detail, then bottlenecks and scaling.
  4. Use the Advanced Topics tab to look up any building block you reached for but could not explain clearly, such as consistent hashing, Bloom filters or the saga pattern.
  5. Draw the final design in the Diagram Tool from the Whiteboarding tab, and tick the steps and Core Concepts you handled well.

Worked examples

What a 5-minute data loss tolerance means

Input

A payments service at 800 transactions a second; the bank has set a maximum data loss of 5 minutes for this critical operation

Output

Up to 240,000 transactions inside the tolerance window, so replication lag must stay well under 5 minutes

800 x 300 seconds is 240,000. The tolerance is the bank's own figure, not one APRA sets, but it turns straight into an engineering target: asynchronous replication with lag monitored and alerted long before it reaches 5 minutes, plus a way to reconcile anything lost. If the interviewer gives you a tolerance of zero, that is a case for synchronous replication.

Synchronous writes from Sydney to Melbourne

Input

Assume a 1,000 km fibre path between the two regions; light in fibre covers about 1 km every 5 microseconds

Output

About 5 ms one way and 10 ms per round trip, so one strictly sequential writer manages about 100 commits a second

Every synchronous commit waits at least one round trip, before any processing time. That is why designs either batch and parallelise writes or keep synchronous replication within one region's availability zones and copy to the second region asynchronously. AWS, for example, lists Sydney and Melbourne as its two Australian regions.

Notices a disruption can trigger

Input

A critical operation is down beyond its time tolerance and an outage of this size is likely to have a material impact

Output

APRA notified within 24 hours of a disruption outside tolerance; a material operational risk incident within 72 hours

Both clocks run from the entity becoming aware. For the design, that means monitoring that measures each critical operation against its own tolerance, not just CPU and error rates, so the people who must notify APRA know which line has been crossed.

Background and insight

System design rounds are open-ended on purpose. There is no single right answer, so the interviewer is scoring how you get to one: whether you pin down requirements before drawing boxes, whether your numbers are in the right order of magnitude, whether you can go deep on one component, and whether you name trade-offs instead of presenting one option as free. Seniority shows mostly in the last two.

Back-of-the-envelope numbers anchor everything. A day has 86,400 seconds, so 100 million requests a day is about 1,160 per second on average, and peak is commonly two to three times that. A 7-character Base62 short code gives 62⁷, about 3.5 trillion, combinations. A single SSD-backed database node can usually serve thousands of simple indexed reads per second, so 1,000 QPS rarely needs sharding, while 100,000 usually needs caching or partitioning.

The CAP theorem is often stated as "pick 2 of 3", which is misleading. Network partitions happen whether you choose them or not, so the real decision is what the system does during a partition: refuse some requests to stay consistent, or keep answering and risk stale reads. PACELC extends this with the latency versus consistency trade-off that applies even when the network is healthy, and it is a better lens for most interview discussions.

Common use cases

  • Preparing for a senior engineering loop

    Practise one prompt every few days with the four-step framework until the timing feels natural.

  • Filling knowledge gaps

    Use the Advanced Topics list as a checklist of building blocks you should be able to explain in two sentences.

  • Running mock interviews

    As an interviewer, pick a prompt and use its focus areas as the probes for the detailed design step.

  • Reviewing real designs

    Apply the same requirements, estimates and trade-offs structure when reviewing a design doc at work.

Tips and limitations

  • Translate the three CPS 230 tolerances into recovery time, recovery point and degraded-mode service, and write them under non-functional requirements.
  • Draw the dependencies on other providers. CPS 230 includes disruptions at material service providers in the severe but plausible scenarios entities must test.
  • Keep both copies in Australia unless told otherwise, and say what that costs: two regions roughly 1,000 km of fibre apart mean asynchronous replication between them.
  • For systems with personal information, add a way to work out quickly whose data was involved. The OAIC treats the 30-day assessment period as a maximum, not a target.

Sources and further reading

  1. Prudential Standard CPS 230 Operational Risk Management (amended version released 30 April 2026), Australian Prudential Regulation Authority. Amended standard commencing 1 July 2026 (the original took effect on 1 July 2025): critical operations, the three tolerance levels and the 24 and 72-hour notifications.
  2. Operational risk management, Australian Prudential Regulation Authority. Current CPS 230 page, including the targeted amendments released in 2026.
  3. Part 4: Notifiable Data Breach (NDB) scheme, Office of the Australian Information Commissioner. Serious harm, the 30-day assessment and notifying as soon as practicable.
  4. AWS Regions, Amazon Web Services. Sydney (ap-southeast-2) and Melbourne (ap-southeast-4).

Frequently asked questions

What is CPS 230 and why does it come up in Australian interviews?

CPS 230 is APRA's prudential standard on operational risk management for banks, insurers and super funds, which commenced on 1 July 2025. For each critical operation an entity sets tolerance levels for the maximum disruption time, the maximum data loss and the minimum service levels under alternative arrangements. Those map directly onto the recovery and resilience part of a system design answer.

Which operations count as critical under CPS 230?

Unless it can justify otherwise, a bank must treat payments, deposit-taking and management, custody, settlements and clearing as critical; an insurer, claims processing; a super fund trustee, investment management and fund administration; and every regulated entity, customer enquiries and the systems that support critical operations.

How fast must APRA be told about an outage?

As soon as possible and within 24 hours if a critical operation is disrupted outside its tolerance, and within 72 hours of becoming aware of an operational risk incident likely to have a material financial impact or a material impact on critical operations. Information security incidents reported under CPS 234 do not need a second report under CPS 230.

What does the Notifiable Data Breaches scheme mean for a design?

Organisations covered by the Privacy Act 1988 must notify affected people and the OAIC of a breach likely to cause serious harm. Where a breach is suspected, the OAIC expects all reasonable steps to finish the assessment within 30 days, and treats that as a ceiling. Audit logs you can query by person make that achievable.

Should I replicate synchronously between Sydney and Melbourne?

Usually not for every write. At roughly 10 ms per round trip on a 1,000 km fibre path, each synchronous commit pays that latency, which limits sequential throughput. A common answer is synchronous replication across availability zones in one region and asynchronous copies to the other, with the data loss tolerance setting how much lag you can accept.

Is there an in-depth guide to System Design Interview?

Yes. “System Design Interviews: A Four-Step Method with Real Numbers” covers the background, how it works, common pitfalls and worked examples. Read the full guide.