Status and error reference

Read the object's current reason and message together with its phase. A phase alone does not identify the cause, and a stale screenshot is not live evidence.

Signal	Meaning	Next check
Starting / not ready	Runtime reconciliation, download or initialization is incomplete	Model Logs and current Pod/events
WaitingForPod	No model Pod exists yet	Operator/KubeAI reconciliation and admission
ModelPodCreationStalled	The no-Pod grace period expired; reconciliation continues	Model ownership, scheduling and controller errors
WaitingForRuntime	A required approved runtime descriptor/image is not available	Runtime module and image promotion
GpuSharingBlocked	Sharing cannot admit/reconcile the requested allocation	GPU sharing state and claim/device-plugin status
Degraded	An actionable dependency/runtime error is reported	First error in the status and matching logs
Removing	Runtime cleanup is still progressing	Terminating Pods and GPU allocation release
Disabled	The saved activation is stopped	Start it when intended; do not delete it to resume
RealtimePermissionDenied	The operator cannot manage a required runtime object	Matching namespaced RBAC and controller release
ModelPodRecoveryExhausted	Bounded terminal-Pod recovery attempts were used	Original error and recovery history

Host operations have a separate lifecycle. Succeeded for memory configuration does not mean model validation passed; network RolledBack confirms restored configuration, not every external connection. See host management and network recovery.

The operator's status contract and runtime CRDs contain further details.