diff --git a/k8s-ng/core/postgresql/cluster.yaml b/k8s-ng/core/postgresql/cluster.yaml
index a8e28f0..46bfe85 100644
--- a/k8s-ng/core/postgresql/cluster.yaml
+++ b/k8s-ng/core/postgresql/cluster.yaml
@@ -2,6 +2,10 @@
 # own node with a node-local volume. Replication is asynchronous (nodes sit in
 # different regions), so a failover can lose the last few transactions.
 #
+# Layout: the primary runs on the music node (the powerful one), the replica on
+# ams-1 (Europe). A failover swaps the roles; switch back with
+#   kubectl cnpg promote main <instance on music>
+#
 # Nodes eligible for instances carry the label workload/postgres=true. To move a
 # replica: label the new node, unlabel the old one, then recreate the instance
 # (kubectl cnpg destroy main <N>); the operator re-clones it from the primary.