System Design. The interview nobody else can run.
Forty-five minutes on a shared whiteboard with an interviewer that asks why and changes a constraint at minute 35 to find out whether you revise or defend. Async platforms cannot do this because design is a conversation. Human services can, at $350 to $500 a session.
What forty-five minutes looks like
The structure comes from the Interviewer’s Guide and it is visible to the candidate as a progress rail, because a hidden clock produces anxiety rather than signal.
| Minute | Phase | What is being observed |
|---|---|---|
| 0 to 5 | Setup and tone | The interviewer states the problem and the time structure out loud |
| 5 to 20 | Requirements | Does the candidate establish functional and non-functional requirements before drawing, and do they surface scale and consistency constraints without being asked |
| 20 to 35 | Deep dive | Depth under three follow-ups in any direction. This is where most of the score is decided |
| 35 to 45 | Critique and constraint change | The distinguishing phase. A new constraint is introduced. A strong candidate revises the design and says what it costs. A weaker one defends the original answer |
The hard guard at minute 25A session still stuck in requirements at minute 25 has gone off the rails, and the interviewer moves it on rather than letting the candidate run out the clock on scoping. That guard is in the book, and it is one of the reasons a calibrated interview beats an enthusiastic one.
The whiteboard is not a picture
What the candidate draws is exported as a node and edge graph, so the scoring reads meaning rather than pixels.
What the canvas gives you
- An architecture stencil set: service, store, queue, cache, client, edge, group
- Typed edges: naming a connection as sync request, async publish, replication, read path or write path, or leaving it untyped on purpose
- A requirements pane, with functional, non-functional, assumptions and out-of-scope tracked separately from the diagram itself
- An interviewer who asks about the design as you talk it through, the same conversation a whiteboard interview has always been
- A napkin-math pane for capacity estimates, kept separate from the diagram
What it deliberately is not
- A drawing test. Neatness is not scored and never will be
- A free-form canvas with no vocabulary, which produces diagrams no two reviewers read the same way
- An image. Nothing here is scored by looking at a picture
- Usable on a phone. The interview refuses to run on one rather than producing a bad score from a cramped canvas
How it is weighted
Problem Decomposition and Architectural Judgment carry the most weight here. Architectural Judgment leads only at Lead / Staff+, where its share rises to 0.35, and that shift is a large part of the difference between a senior and a staff interview.
| Dimension | Junior / Mid | Senior | Lead / Staff+ |
|---|---|---|---|
| Problem Decomposition | 0.30 | 0.30 | 0.30 |
| Architectural Judgment | 0.25 | 0.30 | 0.35 |
| Technical Depth | 0.25 | 0.25 | 0.20 |
| Communication | 0.20 | 0.15 | 0.15 |
Send one to yourself first.
Forty-five minutes, and you get the same report a candidate would generate.
The other interview types: Technical Screen · Coding · Data Science