add low-level alerts and runbook (if appropriate), for example:

• requests being rejected by the apiserver, or concurrency limit is being exceeded in priority and fairness.
• requests timing out
• aggregated APIs are slow or failing
• writing of audit log  is failing
• apiserver --> etcd slowness
• serializing a response and writing to http response body is slow

these low level alerts have defined scopes and are more actionable. if these alerts fire at the same time as KubeAPIErrorBudgetBurn then we have a pretty good idea of what is contributing to the budget burn.