lib/clusterListModel.ts derives lifecycle categories, counts and filters from the same organization snapshot; its tests pin missing-data and self-hosted semantics. A running lifecycle is not a monitoring-health assertion. Regions combine resolved plan labels with available non-retired instance labels, deduplicating matches, because organization responses need not expand instances; the list resolves opaque ids through the existing organization-scoped region catalog once, without polling. Missing labels remain unreported without discarding other reported instance regions. Running classification uses the same activeClusterStatuses rule as card navigation; UPDATED remains attention until RUNNING is reported. The model still derives instance count and version when instance data is supplied, but the card no longer renders them: the organization response does not currently expand instances, so those two cells only ever showed a dash, and the Regions cell (real plan labels) went with the row. The list adds no per-card detail polling. Existing ClusterProgress retains its lifecycle polling behavior.
SystemStatus notices have no cluster association and remain in the cloud-wide notification banner/center. Per-cluster monitoring incidents and scheduled maintenance need a scoped backend contract before the list can attribute them. The UI preserves ClusterCard's permission checks and lifecycle actions.
The upsert page keeps a single form owner for create, edit, and version-only flows. Details submit remains inside the HTML form; payment review uses its existing click handler outside that form. The shared price summary uses the existing price calculations on both steps and follows the fields on narrow screens. cluster-form.anon.spec.ts pins request payloads, partial-upgrade retry, billing transitions, and responsive bounds.
The form terminates a cluster only after creating its replacement: Try Again on a failed cluster's card opens a
create prefilled from it, and once that create succeeds the form terminates the original if it's still FAILED.
sourceClusterId names that original, but the form also fills it with the edited cluster's own id on an edit, and
saved form values can bring it back after a billing redirect. So onClusterSavedCallback gates the terminate on
creating, not on sourceClusterId alone. Without that gate, an update that failed straight away made an edit delete
the cluster it was editing.