Governance Domains
Resilience
Coverage ratios over five configuration signals, gated so a rating is issued only when both survival and recoverability were observable.
Weight: 0.3.
The signals
| Signal | Resource family |
|---|---|
| Databases configured across multiple availability zones | RDS |
| Databases retaining automated backups | RDS |
| Auto Scaling groups spanning two or more zones | Auto Scaling |
| Load balancers spanning two or more zones | Load balancing |
| Object storage with versioning enabled | S3 |
Each is a coverage ratio over real resources. A signal with no in-scope resources is recorded and excluded from scoring, so an account that runs no databases is not penalised for it.
The gate
Two governance concerns, not five signals:
| Concern | The question |
|---|---|
| Availability | Do workloads survive a failure? |
| Recoverability | Can data be recovered? |
Both must be evidenced before the domain is scored. A rating that can see only one of them is half a resilience rating.
Why the gate is not a signal count
The five signals are neither independent nor evenly spread. Databases emit two of them from a single population, so counting signals would let a database-only estate satisfy a breadth test it has not met, while an estate spanning storage and load balancing failed one it had.
Resources are counted once per family, so a service exposing two signals from one population does not count twice.
Minimum population
At least five distinct applicable resources, counted per family.
What the collector does not observe
Point-in-time recovery on managed NoSQL, AWS Backup plans, EBS snapshots, EFS, Aurora, container services, cross-region replication, and standalone compute instances.
For a predominantly serverless estate its coverage of the real resilience surface is close to zero. That is published rather than absorbed. See Coverage limits, domain by domain.