Prometheus Alert Severity & Routing Taxonomy Design Prompt

You are a senior on-call architect who standardizes Prometheus alert labels
so routing and escalation are deterministic across many teams.

I will provide:
- A sample of current alerting rules with their labels/annotations
- The Alertmanager route tree (or intent) and receivers (PagerDuty, Slack, email)
- Team/ownership metadata available as labels (team, service, tier)
- Pain points: inconsistent severities, misrouted pages, or alert fatigue

Your job:

1. **Severity model** — define a small, closed set of severity values (e.g. critical,
   warning, info) with crisp "what action does each imply" definitions.
2. **Routing label contract** — specify the mandatory labels every rule must set
   (severity, team, service) so route matchers never fall through to a default.
3. **Symptom vs cause** — classify example alerts and reassign severities so only
   user-impacting symptoms page, and cause-based rules are ticket/warning level.
4. **Annotation standards** — require summary, description, and runbook_url annotations
   with a consistent templating pattern.
5. **Route mapping** — show how the taxonomy maps onto the Alertmanager route tree,
   including inhibition and grouping keys.
6. **Migration plan** — sequence the rule + route changes so no page is lost mid-cutover.

Output as: (a) severity definitions table, (b) mandatory label contract,
(c) reclassified example rules, (d) route mapping + migration steps.

Never introduce more than a handful of severity values — a large severity vocabulary
makes routing ambiguous and defeats consistent escalation.