diff --git a/tests/discourse/compose.ccci.yml b/tests/discourse/compose.ccci.yml
index 1053580..9d8d357 100644
--- a/tests/discourse/compose.ccci.yml
+++ b/tests/discourse/compose.ccci.yml
@@ -20,6 +20,19 @@ version: "3.8"
 #      ships 20m, so this overlay is idempotent on the head (it persists untracked across the checkout).
 # Both changes are namespace/grace-only: identical image content, a healthy check still marks healthy
 # immediately → NO assertion is weakened and no defect is masked.
+#
+#   3. PREPULL-ENABLE (compose-validity): the published recipe (BOTH 0.7.0 base AND the PR head) ships
+#      `sidekiq.depends_on: [discourse]`, but the main service is named `app` — `discourse` is an
+#      UNDEFINED service, so `abra app config --images` (the harness prepull step) returns
+#      `invalid compose project` (rc=15) and prepull is SKIPPED → the 2.4GB discourse image is pulled
+#      INLINE during `abra app deploy`, eating the convergence budget on the 7-GiB single node and
+#      pushing the slow Rails cold boot past the deploy timeout (full4 base-deploy timed out at 2400s).
+#      This overlay overrides sidekiq.depends_on to the real service `app` so the merged compose is
+#      VALID → prepull resolves+pre-pulls the images before deploy → reliable convergence. This is a
+#      NO-OP at runtime: `docker stack deploy` (swarm) IGNORES depends_on entirely, so correcting a
+#      dangling lint-only reference to the real service changes NOTHING swarm does — it masks no defect
+#      and weakens no test; it only unblocks the image pre-pull. (The same dangling ref exists upstream;
+#      filed as a recipe nit, but the fix here is the immutable-published-base + harness-prepull path.)
 services:
   app:
     image: bitnamilegacy/discourse:3.3.1
@@ -27,3 +40,5 @@ services:
       start_period: 20m
   sidekiq:
     image: bitnamilegacy/discourse:3.3.1
+    depends_on:
+      - app
diff --git a/tests/discourse/recipe_meta.py b/tests/discourse/recipe_meta.py
index f316cd8..2af8ff6 100644
--- a/tests/discourse/recipe_meta.py
+++ b/tests/discourse/recipe_meta.py
@@ -6,7 +6,8 @@
 # app is actually serving (the canonical "is discourse up" signal — NOT "/", which may redirect to setup).
 HEALTH_PATH = "/srv/status"
 HEALTH_OK = (200,)
-DEPLOY_TIMEOUT = 2400  # slow Rails cold boot (15-25min); matches the EXTRA_ENV TIMEOUT below
+DEPLOY_TIMEOUT = 3600  # slow Rails cold boot (15-25min) on the 7-GiB single node; bumped 2400→3600 for
+# headroom after full4's base deploy timed out at 2400s (RAM/CPU-constrained boot + image re-pull).
 HTTP_TIMEOUT = 1200
 
 # Slow-cold-boot handling: the recipe-PR (recipe-maintainers/discourse#1) bumps the app healthcheck
@@ -33,7 +34,7 @@ HTTP_TIMEOUT = 1200
 CHAOS_BASE_DEPLOY = True
 UPGRADE_BASE_VERSION = "0.7.0+3.3.1"
 EXTRA_ENV = {
-    "TIMEOUT": "2400",
+    "TIMEOUT": "3600",  # abra's internal convergence wait; matches DEPLOY_TIMEOUT (slow Rails boot headroom)
     "COMPOSE_FILE": "compose.yml:compose.ccci.yml",
 }