staff-engineering · staff

Architecture Reviews Without Theater

Quick answer

An architecture review is a structured conversation that asks: Does this design meet the quality-attribute scenarios that matter, and what risks remain? Staff engineers facilitate reviews to find problems early, record decisions, and protect teams from unexamined blast radius—not to humiliate authors or collect prestige.

Why this matters

Learning objectives

  1. Prepare a review packet (problem, scenarios, design, risks).
  2. Facilitate time-boxed, inclusive critique.
  3. Separate blocking issues from suggestions.
  4. Close with decisions and owners.
  5. Avoid review anti-patterns.

Explain like I am 5

Before the school play, someone checks that the stage can hold the cast and the exits are clear—not to be mean, but so the show is safe.

Mental model

flowchart TD
  Prep[Packet: problem + QAS + design] --> Review[Facilitated session]
  Review --> Issues[Blocking vs non-blocking]
  Issues --> Decision[Accept / revise / reject]
  Decision --> ADR[ADR + follow-ups]
RoleResponsibility
AuthorClear packet; honest risks
FacilitatorTime, inclusion, decisions
ReviewersScenario-based critique
ScribeDecisions and actions

Core concepts

Packet before people

Send materials ≥24–48h ahead: problem, quality-attribute scenarios, context diagram, key flows, failure modes, open questions. Surprise reviews reward the loudest voice.

Scenario walkthroughs

Pick 3–5 critical scenarios and walk the design through each. This beats free-form “I don’t like Kafka.”

Blocking vs non-blocking

Blocking: safety, data loss, compliance, SLO impossible. Non-blocking: style, preferred libraries, future-nice-to-haves. Record both; only blocking holds ship.

Decisions in the room (or async SLA)

Leave with accept/revise/reject and named owners. “We’ll think about it” is not a decision.

Psychological safety

Critique the design, not the person. Staff facilitators shut down status games and invite quieter experts (security, SRE, data).

Worked example

Reviewing a multi-region payments write path:

  1. Walk partition and dual-write scenarios.
  2. Blocking: no reconciliation strategy for partial commits.
  3. Decision: revise with outbox + reconciliation job; re-review delta in one week.
  4. ADR drafted from the decision; security follow-up ticket owned.

Trade-offs

Heavy review boardNo reviews
Slow innovationSurprise incidents
Strong governanceLocal speed

Staff default: lightweight reviews for high blast radius; skip for well-paved patterns.

Failure modes

ModeMitigation
Slide-only theaterRequire scenarios + risks
Design by committeeFacilitator + decision owner
Late drive-by nitsTime-box; parking lot
Rubber stampRotate independent reviewers
No follow-upAction list with dates

Interview mode

Skeleton: “I run architecture reviews as scenario-driven risk discovery—prepared packets, blocking vs non-blocking issues, and explicit decisions that become ADRs.”

Knowledge check

Whether the design meets critical quality-attribute scenarios and what risks remain

Only whether the slides look polished

Who has the highest title in the room

Avoiding any written follow-up

Related

By Shubham Jain

All articles

Shubham Jain · Learning Lab