lockstep
3 min readNaman Jain

Architecture review template: checklist, meeting, and outcome

A free architecture review template: a checklist for security, data, and operations, how to run the meeting, and how to record the outcome.

A design review runs 90 minutes. Six senior engineers raise good points, the author takes notes, and everyone leaves. Nobody says whether the design is approved. Two weeks later, the team is building it, half the feedback is lost, and one reviewer is surprised the database choice they objected to went ahead.

A review is only as useful as its outcome. This template keeps it short and ends with a clear result, drawing on Malte Ubl's Design Docs at Google, checked on 16 November 2026.

What a review is for

Ubl describes the primary value of design reviews as an opportunity to bring the organization's combined experience into a design. He also warns that reviews can become a trap of overhead, and that waiting for review meetings can slow development; engineers can reduce that by seeking the most important feedback directly.

Light or heavy

Lightweight reviewFormal review
HowShare the doc; discuss in commentsMeeting with senior engineers
WhenMost designsHigh-risk, costly, or contested designs
SpeedDaysDepends on meeting schedule
OutputResolved comments, approver sign-offRecorded outcome and decisions

Default to lightweight. Escalate to a meeting when comments can't resolve a disagreement or the risk is high.

The review checklist

## Problem and scope
- [ ] The problem and goals are clear; non-goals are stated
- [ ] The design solves the stated problem, not a bigger one

## Design
- [ ] Alternatives were considered, with reasons for rejecting them
- [ ] It fits existing systems and conventions, or explains why not
- [ ] Interfaces and data ownership are clear

## Cross-cutting concerns
- [ ] Security: auth, secrets, attack surface
- [ ] Privacy: what personal data, retention, access
- [ ] Data: schema changes, migrations, backfills
- [ ] Reliability: failure modes, timeouts, retries
- [ ] Operations: monitoring, alerts, runbooks, on-call impact
- [ ] Cost: infrastructure and ongoing

## Delivery
- [ ] Rollout plan, with feature flags where useful
- [ ] Rollback plan
- [ ] Open questions have owners and dates

## Decisions
- [ ] Key decisions are listed, with reasons
- [ ] Decisions that should outlive the doc are recorded as ADRs

Running a formal review

  1. Send the design doc 2 days ahead. Reviewers read before the meeting.
  2. Start with the questions the author needs answered, not a full presentation.
  3. Go through the checklist on the risky parts only.
  4. Name a decider. One person states the outcome.
  5. End with the outcome in the last 5 minutes:
## Review outcome: [Design name]
Date: [Date]   Decider: [Name]   Reviewers: [Names]

Outcome: Approved / Approved with changes / Another round needed

Required changes:
- [Change, owner, due]

Decisions:
- [Decision and reason]

Follow-ups:
- [Action, owner, due]

Architecture review boards

Ubl recalls that when Google was smaller, designs went to a single mailing list for senior engineers, which helped create a uniform design culture, but became infeasible as the engineering team grew. A standing review board has the same trade-off: consistency, at the risk of a queue. Keep its scope to designs that are hard to reverse, and let everything else go through lightweight review.

Record what was decided

The review's value disappears if its decisions live only in the meeting. Put the outcome at the top of the design doc, and record decisions that will matter later as architecture decision records, so the engineer, or AI coding agent, changing this system next year knows what was agreed and why.

Where Lockstep fits

Lockstep captures decisions like these from design docs, Slack, and pull requests, routes each to its owner for approval, and briefs Claude Code sessions on the ones that apply to the code being changed.

See how it works.

Sources checked 16 November 2026

For the document being reviewed, see the technical design document template. For proposals with a comment period, see the engineering RFC template.

Frequently asked questions

What is an architecture review?
A review of a proposed design by engineers outside the immediate team, before it's built. Malte Ubl's account of design docs at Google describes its main value as the chance to bring the organization's combined experience into a design.
Does every design need a review meeting?
No. Ubl describes a lightweight form, sharing the doc with the team and discussing in comments, and a heavy form, a formal meeting with senior engineers. Meetings can slow work while teams wait; use them for high-risk or contested designs.
What should an architecture review check?
Whether the problem and goals are clear, alternatives were considered, the design fits existing systems, and cross-cutting concerns are covered: security, privacy, data, reliability, operations, cost, and how it will be rolled out and rolled back.
What is an architecture review board?
A standing group of senior engineers or architects who review significant designs on a regular schedule. It can create a consistent design culture, but at scale it can become a bottleneck if every change must wait for it.
What should come out of an architecture review?
An explicit outcome, such as approved, approved with changes, or needs another round, plus the decisions made and the conditions attached, recorded where the team and future engineers can find them.

Keep reading