Operation In Practice

Operation keeps the system reliable, observable, and cost-effective. The guide covers what each operational practice is. This page covers how to apply them.

Relationship Map

Operation is a continuous loop. Production data flows back into Direction and Delivery:

Monitoring (3 tiers) ──► Incident Response ──► Post-Mortem ──► Feedback
      │                                                          │
      │                                                          ├── Direction: user behaviour changes Missions
      │                                                          ├── Delivery: incidents add Stories or tech debt
      │                                                          ├── Operation: cost trends trigger FinOps
      │                                                          └── Operation: SLO breaches shift to reliability
      │
      ├── On-call rotation (primary + secondary)
      ├── SLOs + error budget (features vs reliability)
      ├── Production Readiness checklist (new services)
      ├── Cost management (once per cycle, 15 min)
      └── Friction budget (estimated at retro)

Default Path + Branching

Default — in production:

Monitoring (3 tiers) + On-call rotation + SLO tracking
  → Incident response when needed
  → Post-mortem after significant incidents
  → Cost review once per cycle
  → Friction budget estimate at retro

Pre-PMF or no production users:

Skip monitoring infrastructure. Watch for pain manually.
Skip on-call rotation (everyone is on-call).
No SLOs. Measure latency when something feels slow.
Skip cost management (cost is negligible).

Incident-heavy period:

Tighten alerting → Reduce toil → Add runbooks →
Post-mortem every incident → Action items → Process improvement

Cost growing faster than users:

Add cost review → Identify largest driver → Optimize or cap →
Track cost per user → Act when one service exceeds 20% of budget

Role Map

Technique Table

Procedure Table

SLO Design

Do not track more than 3 SLOs per service. Use the error budget: full → ship features. Low → fix reliability.

Incident Severity

Write a runbook for every incident type that repeats twice. If the same incident happens twice without a runbook, that is a process gap.

On-Call Design

Production Readiness Checklist

When to skip: Internal tools with fewer than 10 users. Short-lived experiments.

Cost Review

Duration: 15 minutes. One person reviews. Not a meeting.

Artifact Table

Skip Conditions

Alert on symptoms, not causes. If an alert does not require a human to act, remove it. If a practice takes longer than the value it creates, remove it.