SA-201c · Module 3

Running Effective Design Reviews

3 min read

A design review should produce decisions. Not opinions, not "alignment," not vague comfort that "the approach looks good." Decisions — documented, with rationale, with owners, and with conditions for revisiting. The design review is the quality gate where architecture is tested before it is built. A well-run review catches the anti-pattern, the missed requirement, and the load-bearing assumption that the architect did not realize was load-bearing.

  1. Pre-Read Distribution (48 Hours) Distribute the architecture document — Level 2 diagrams, ADRs for open decisions, and a list of questions that need answers — 48 hours before the review. Attendees who arrive having read the document produce decisions. Attendees who arrive cold produce questions that the document already answers.
  2. Questions-First Format Start the review with the open questions, not a presentation of the architecture. "Here are three decisions we need to make. Here are the options and trade-offs for each." This orients the room toward output from the first minute. If nobody has questions about the architecture itself, the pre-read worked.
  3. Decision Documentation Document every decision in the meeting as it is made. Decision, rationale, owner, and review trigger. If a decision cannot be made, document what information is needed and who will provide it by when. The meeting notes are ADRs in draft form — finalize them within 24 hours.
  4. Time-Boxing Design reviews have a fixed duration. Sixty minutes for a standard review. Ninety minutes for a complex review. When the time expires, any undecided items are carried forward with a decision owner and a deadline. Open-ended reviews produce conversation without convergence.

Do This

  • Distribute the architecture document 48 hours before the review
  • Start with open questions that need decisions — not a walkthrough of the document
  • Document every decision live with rationale, owner, and review trigger
  • Time-box the review — undecided items get owners and deadlines, not more meeting time

Avoid This

  • Present the architecture cold to an unprepared audience
  • Let the review become a walkthrough of every diagram with no decision structure
  • End the review with "we are aligned" without specifying what was decided
  • Allow the review to run until everyone has spoken — convergence, not completeness, is the goal