Running a backend design round
Finish the system's outline, then spend depth on the choices that need evidence.
Predict first
Before you read: predict
In a backend catalog round, the agreed capabilities are browsing and price updates, and the board names Product, Category and Price. The candidate has sketched a service and a database box.
Which artifact should lead the next part of the conversation?
The 30-second version
Run the backend round through agreed scope, core entities, the caller contract, a complete service blueprint, and focused deep dives. The API precedes the backend blueprint in this framework. Reserve a closing checkpoint. Name the hard parts and explain limits yourself. Our rehearsal chooses two dives; that is a practice constraint, not a universal interview law. [S1–S4]
Finish the system's outline, then spend depth on the choices that need evidence.
How it works
Step 1 of 8
se-requirements-nfrs. [S1, S4]Analogy: A bespoke bookbinding commission
A bookbinder reviewing a commission first agrees how the book will be used. The binder names the text block, cover and spine, agrees the dimensions at their joins, then sketches the entire binding. Detailed trials can focus on a difficult hinge and an unusual fastening. A closing note distinguishes the promised use from conditions the prototype has not tested.
| Commission review | Backend round |
|---|---|
| Intended use | Agreed requirements |
| Named physical parts | Core entities |
| Dimensions at handoffs | Interface contract |
| Whole binding sketch | High-level design |
| Selected joint trials | Focused deep dives |
| Unproven use conditions | Explicit design limits |
The analogy concerns the order of a review, not the mechanics of binding a book. Physical joints do not model network contracts exactly. It also cannot establish which database, protocol or failure policy a system needs. Those decisions require their own technical evidence. The mapping helps keep a complete outline visible while one difficult detail receives attention.
Try it
Answer first, then check. Nothing is saved.
Question 1 of 5
In an eight-minute review, a price change reaches 40,000 open catalog pages 9 seconds late against a 1-second target, and two concurrent checkouts both reserve the last unit of a flash-sale item. Category-filter latency meets its target; cache-hit rate is stable. The candidate has two deep-dive slots.
Which pair should lead?
Question 2 of 5
This practice round lasts 45 minutes. At minute 27 the candidate has completed the catalog blueprint. The house rule is to protect a closing reserve, and the facilitator has selected two deep dives. No rounding is needed.
Following the rehearsal schedule, how many minutes does each deep dive receive?
Question 3 of 5
The two planned dives concern feed reconnection and recovery of an acknowledged price update.
Which treatment shows practical depth?
Question 4 of 5
The candidate has agreed requirements, entities and API calls. The whiteboard contains service and storage boxes, and the next card is queue retry policy.
Which card should the facilitator place before it?
Question 5 of 5
The walkthrough has a working browse path. During the board review, the candidate notices that a reconnect after a dependency fault has not been explained.
Which closing contribution is most useful?
Go deeper
Pro tip
Authored communication suggestion: keep the same names for services and resources as the whiteboard develops. Use a small legend if you change a name. This is a rehearsal aid, not a sourced interview requirement or an assessment key.
Related lessons
Checked against the tools and docs as of 2026-09-29.