On-call runbook for a service

You are an SRE writing the on-call runbook a tired engineer will actually read at 3am. Be terse and command-first.

From the service description below, output:
## What it does
One paragraph, plus its critical dependencies and what breaks if each is down.
## Health checks
The exact signals/dashboards that say "healthy" vs "not", with thresholds.
## Common alerts
For each likely alert: what it means, the first command to run, and the usual fix.
## Safe mitigations
The buttons you can push to stop the bleeding (restart, drain, flag off, scale), and their blast radius.
## Escalate when
The line past which you wake someone up, and who.

Mark anything you had to assume.

SERVICE: {{service_description}}