Everyone's rushing to adopt Claw runtimes because they're trendy. They're ignoring the operational debt.
Public scorecards are useless. They're marketing checklists. You need to evaluate based on your own failure modes. Here's the core of what I'd look at in a real rubric:
```yaml
evaluation_criteria:
- criterion: "Secret injection at runtime"
test: "Deliberately rotate vault token during a canary deployment. Does the runtime crash, hang, or leak?"
weight: high
- criterion: "Configuration drift detection"
test: "Deploy with Ansible, then manually edit a config file in the container. Does the runtime alert or self-heal?"
weight: medium
- criterion: "Chaos readiness"
test: "Simulate a downstream dependency 500 error. Are metrics/logs actionable for a postmortem, or just 'error occurred'?"
weight: high
```
Forget "easy to integrate." Measure how it dies.
Don't panic, have a rollback plan.