In the single-cluster overview dashboard, we have two stats that show the number of healthy nodes and whether that's all the nodes in the cluster. This is calculated by combining cbnode_healthy and cbnode_cluster_membership from the Exporter.
However, when I made the dashboard I misinterpreted the latter metric: it treats a node that's failed over as not a member of the cluster, meaning that e.g. with three nodes and one failed-over it'll show that 2/2 are healthy and thus everything is fine.
Look into alternate ways of implementing this, or if there are none, remove the "All Nodes Healthy" metric.
On a similar though less damaging note, "Unhealthy Nodes" can be misleading when there is no data (for example, misconfigured exporter) as it'll show 0.