diff --git a/playbooks/deploy-skaldsong.yaml b/playbooks/deploy-skaldsong.yaml
index 9094e82..c6bc254 100644
--- a/playbooks/deploy-skaldsong.yaml
+++ b/playbooks/deploy-skaldsong.yaml
@@ -112,6 +112,12 @@ verify:
     shell: curl -sf -o /dev/null http://localhost:{{ host_port }}/health
     changed_when: "false"
 
-  - name: skaldsong container running + healthy
-    shell: docker ps --filter name=^/skaldsong$ --format '{{.Status}}' | grep -q 'healthy'
+  - name: skaldsong container running
+    # Grep for '^Up' (not 'healthy') — verify runs immediately after the
+    # deploy step finishes, which is well before the container's
+    # start_period (30s) elapses. The 'healthy' state from docker ps is
+    # a delayed echo of the same /health probe verify 1/2 already
+    # confirms, so racing the start_period here is bogus. Keep this
+    # check to catch the "container exited entirely" case.
+    shell: docker ps --filter name=^/skaldsong$ --format '{{.Status}}' | grep -q '^Up'
     changed_when: "false"