Architecture Review Practice
Running a design review that improves the design rather than performing gatekeeping, through timing, framing and the questions asked.
Reviews go wrong in a consistent way: scheduled too late to change anything, attended by people seeing the system for the first time, focused on compliance with standards rather than on whether the design will work, and experienced by the team as an obstacle.
The changes that fix it. Timing: review when the design is formed but not committed — early enough that a finding can change the shape, late enough that there is something to examine. A review the week before build starts is theatre.
Preparation: material circulated in advance and read, so the session is spent on discussion rather than on presentation. A review where the first forty minutes is a walkthrough has no time left for the part that adds value.
Framing: the reviewers' job is to improve the design and surface risk, not to approve. Where a sign-off is genuinely required, separate it from the improvement conversation, because the two produce different behaviour in the room.
Questions over verdicts: "what happens when the downstream is slow rather than down", "how does this behave at ten times the volume", "how will you know this is broken at 3 AM", "what is the rollback". These generate better outcomes than a checklist, and they teach.
The measure of a healthy review culture is that teams request reviews for hard problems rather than scheduling them because policy requires it.