How a system design interview works
It is a design review with a stranger, on a clock, scored on four signals.
What you will be able to do
- Describe what a design round is and what it is not
- Name the four signals interviewers write evidence against
- Budget a 60 minute slot into phrases with times attached
- Explain how the same prompt is graded differently at mid, senior and staff level
- Open a round with three sentences that show you are driving
Picture the room
You are on a call with an engineer you have never met. They say eight words: "Design a URL shortener." Then they stop talking. There is no spec, no scale figure, no existing system to read. For the next 40 minutes you ask questions, choose components, estimate numbers, and defend trade-offs while they occasionally interrupt.
That framing explains most of the format. The prompt is vague because real projects arrive vague. The interviewer interrupts because real colleagues push back. The clock matters because a design review that never reaches the hard part was a wasted meeting. They are not comparing your diagram to an answer key — they are asking one question: if this person joined my team, would their design reviews be worth attending?
The four signals on the scorecard
Companies word their rubrics differently, but the same four signals show up almost everywhere. Each one is scored with evidence: quotes from the round, moments noted down.
| Signal | Strong | Weak |
|---|---|---|
| Problem navigation | Asks the questions that change the design, states scope out loud, picks the two or three parts worth going deep on | Starts drawing before knowing what matters; spends ten minutes on the login page of a video platform |
| Solution design | A simple design that satisfies every functional requirement, with a request you can trace end to end | Impressive components nothing connects to, or a database nothing ever writes to |
| Technical depth | Chooses an option, names its cost, backs the choice with a number | A list of technology names with no trade-off attached |
| Communication | Thinks out loud, checks in at phase boundaries, treats a hint as a gift | Long silences, or a ten minute monologue that ignores the interviewer |
Communication is the only signal scored in every minute of the round. A clean diagram still fails if the interviewer could not follow the reasoning behind it, because nothing they could not hear can go into the write-up.
Where the hour goes
Most loops run 45 or 60 minutes. A 60 minute slot loses about five minutes to introductions and five to ten to your own questions, so budget for 45 minutes of design. Phase by phase:
| Phase | Time | What you are doing | Mostly scored for |
|---|---|---|---|
| Intro | 0-5 | Names, one line of background, restate the prompt back to them | Nothing yet — but keep it short |
| Requirements | 5-10 | Ask what changes the design; write scope, non-goals and rough scale | Navigation |
| Entities and API | 10-15 | Name the nouns of the system and the calls that touch them | Design |
| High-level design | 15-30 | Boxes, arrows and data flow that cover every functional requirement | Design |
| Deep dives | 30-45 | Two or three hard parts, solved with trade-offs and numbers | Depth |
| Wrap-up | 45-55 | Recap, state the weakest part, ask your questions | Communication |
The bar moves with level
The prompt is often identical across levels. What changes is how much of the work happens without help, and how far past the basic design you get.
| Mid-level | Senior | Staff | |
|---|---|---|---|
| Who drives | You follow the structure; they steer you between phases | You drive the whole round and they mostly react | You drive, and you reshape the prompt when it is framed badly |
| Depth | A working design and solid basics | A working design fast, then real depth in two or three areas | Depth plus cost, operations, migration and team concerns |
| Finding the hard parts | They point at them and you solve them | You name them before being asked | You rank them and say which one matters most for this business |
| Trade-offs | Knows the options exist | Picks one, names the cost, backs it with a number | Knows when the right answer changes as requirements change, and says when |
| Failure | Mentions retries or replicas when asked | Plans for overload and failure unprompted | Talks about blast radius, degradation, and how you would know it broke |
One line worth remembering: mid-level solves the problem they are given, senior finds the hard problem inside it, staff questions whether it is the right problem. A mid-level candidate who delivers a clean design and one good deep dive has done well. A senior candidate with the same round often has not, because nobody saw them lead.
How to open in three sentences
Given "Design a URL shortener" and no other instruction, here is an opening that puts all four signals on the board before you draw anything.
Notice what is absent: no technology names, no diagram, no estimates. Those come once the problem is scoped. Naming Kafka in minute two is not depth — it is a guess dressed as a decision.
What happens after you leave the call
Interviewers work from a short prepared list of probes for each prompt: what happens when this shard gets hot, how does a user read their own write, what if this service goes down for ten minutes. Then they write evidence against each signal, with quotes attached.
- Raise a probe before they ask it. That is the single strongest signal available, because it shows you would catch the problem in a real review.
- Treat a hint as a gift, not a threat. "What if traffic is 100x during a live event?" is not a trap; it is the deep dive they want to watch you do.
- Say it out loud. A good idea you kept in your head cannot appear in the write-up, so from the interviewer's chair it did not happen.
- Read a pushback as steering. "Are you sure that will not be a bottleneck?" usually means they want the partitioning discussion — say what you would check and go there.
Worth remembering
- The prompt is vague on purpose. Turning a vague prompt into a scoped problem is the first thing being scored.
- Four signals: problem navigation, solution design, technical depth, communication. Most rejections come from one weak signal, not a missing technology.
- Of a 60 minute slot you get roughly 45 minutes of design time. Plan for 45, not 60.
- Interviewers probe from a prepared list. Raising one of their probes yourself is worth more than answering it well after they ask.