The Weekly Health Review Format
30 minutes weekly to track every service's health. The agenda, the rotation, and what the team learns that no one-off review surfaces.
Agenda
The agenda is fixed. 30 minutes, four sections, time-boxed. Going over time on any section drops the next one rather than overrunning the meeting.
- 5 min: incidents. Sev 1 and Sev 2 from the past week with action-item status. Sev 3 and Sev 4 are noise at this altitude.
- 10 min: SLO trends. Per-service deltas. Skip services that did not move; spend the time on the ones that did.
- 10 min: burn-rate and projection. Where the SLOs land at quarter end if the current rate continues. This drives the early-warning conversation.
- 5 min: action items. Status update on prior-week items. Owner names attached, deadlines re-stated, blockers surfaced.
Who attends
Attendance is small and purposeful. The room is for people who can act on what they hear; observers belong on the recorded summary, not in the meeting.
- On-call this week. Walks in with the most recent context; leaves with situational awareness for the rest of the rotation.
- On-call next week. Walks in with no context, walks out prepared. Skipping next-week is the most common attendance bug.
- Service owners in red. The responsible team’s lead for any service breaching SLO. Optional otherwise; required when red.
- SRE leadership. One representative who can authorise cross-team decisions or investment in the room. Without them, action items stall.
Output
The output is one page that lives in the wiki. The page is the durable record; the meeting itself is ephemeral.
- One-page wiki summary. Trends, decisions, action items. Searchable archive that future on-calls inherit.
- Remediation plan for red services. Each consistently-red service gets a written plan with an owner and a deadline. The plan is the artefact, not the discussion.
- Cadence is the point. Weekly catches drift before it becomes incident-causing. Skipping two weeks in a row is itself a signal.
- Quarterly format retro. Review the meeting itself once per quarter. Drop sections that stopped earning their time and tune the ones that need more.