Start with facts
A review starts with timeline, impact, detection, mitigation, and recovery. Avoid pretending the team knew the answer earlier than it did.
Prefer system changes
The best action items change the system around the work: alerts, runbooks, safer defaults, rollback paths, ownership, and missing tests.
Good action item: add a queue-age alert with owner, threshold, and runbook. Weak action item: be more careful.
A small review template
## Impact
Who was affected and for how long?
## Detection
How did we learn about it?
## Contributing factors
Which system conditions made it possible?
## Changes
What will reduce recurrence or impact?