Data
Owner, freshness, privacy, dependency shape, authorization, failure behavior.
Quick reference · Print-friendly
Rendering interview playbook
Clarify, partition, assign policies, expose risks, verify.
The answer spine
| Move | Say or draw | What it proves |
|---|---|---|
| 1 · Clarify | Journeys, freshness, privacy, SEO, devices, traffic, failure tolerance | You design from constraints, not fashion |
| 2 · Partition | Stable/volatile, public/private, critical/optional, static/interactive | You see hybrid seams |
| 3 · Assign | Generation, cache, delivery/reveal, client execution | You do not collapse independent axes |
| 4 · Defend | Keys, invalidation, serialization, status, fallbacks, authorization | You understand correctness and operations |
| 5 · Verify | Field cohort, trace, budget, experiment, guardrails, rollback | You close with evidence |
Whiteboard inventory
Cache
Key, TTL, invalidation event, stale window, scope, stampede control.
Delivery
Shell, stream boundaries, fallback geometry, status commitment, abort.
Client
Entry points, transitive graph, hydration/resume, state channel, no-JS path.
Evidence
Field segment, Web Vitals path, cache/origin spans, errors, cost.
Rollout
Canary, feature flag, correctness tests, alert, rollback threshold.
Diagnostic prompts
- Which user and navigation is slow: new, returning, crawler, cold, warm, client transition?
- Which segment dominates: origin, transfer, discovery, browser CPU, interaction, or layout?
- Which evidence would falsify the leading explanation?
- What correctness property could the optimization weaken?
- What happens under cache miss, dependency timeout, partial failure, and retry?
- Which result would make you reverse the architectural choice?
Compact close
“Given these assumptions, I’d partition the page by freshness, privacy, criticality, and interaction. Each region gets an explicit generation, cache, delivery, and client-execution policy. The largest risks are X and Y. I’d validate them with Z field segment and correlated traces, ship behind a canary, and roll back if the correctness or latency guardrail moves.”