diff --git a/sql/05_tiering.sql b/sql/05_tiering.sql
index 531512c..50b93c8 100644
--- a/sql/05_tiering.sql
+++ b/sql/05_tiering.sql
@@ -21,9 +21,6 @@ DECLARE
     v_oid OID;
     v_lsn PG_LSN;
 BEGIN
-    IF to_regclass('wal2delta.tables') IS NULL THEN
-        RETURN NULL;
-    END IF;
     v_oid := to_regclass('lakets_cdf.' || quote_ident(p_shadow_name))::oid;
     IF v_oid IS NULL THEN
         RETURN NULL;
@@ -32,6 +29,13 @@ BEGIN
     FROM wal2delta.tables
     WHERE table_oid = v_oid AND status = 'STREAMING';
     RETURN v_lsn;  -- NULL if no STREAMING row
+EXCEPTION
+    -- wal2delta is a Lakebase-managed schema. If the subsystem is absent
+    -- (undefined schema/table) or this role lacks USAGE/SELECT on it
+    -- (insufficient_privilege), we cannot confirm durability -- fail closed
+    -- so tiering never evicts data we could not verify in Unity Catalog.
+    WHEN insufficient_privilege OR undefined_table OR invalid_schema_name THEN
+        RETURN NULL;
 END;
 $$;
 
diff --git a/sql/13_shadow_sync.sql b/sql/13_shadow_sync.sql
index 9551163..51e686c 100644
--- a/sql/13_shadow_sync.sql
+++ b/sql/13_shadow_sync.sql
@@ -121,6 +121,7 @@ AS $$
 DECLARE
     v_is_ct BOOLEAN;
     v_is_ru BOOLEAN;
+    v_cdf_present BOOLEAN;
 BEGIN
     SELECT EXISTS(SELECT 1 FROM lakets._chronotable_registry
                   WHERE schema_name = p_schema_name AND table_name = p_table_name) INTO v_is_ct;
@@ -130,11 +131,19 @@ BEGIN
     -- enabled on the lakets_cdf schema via Databricks. Without it, the shadow is
     -- created but never streams to Unity Catalog, and tiering stays fail-closed
     -- (nothing is ever evicted). Warn loudly rather than failing, so sync can be
-    -- wired before CDF is turned on.
-    IF to_regclass('wal2delta.tables') IS NULL THEN
-        RAISE WARNING 'Lakebase CDF (wal2delta) is not enabled on this database. The '
-            'shadow will be created but will not replicate to Unity Catalog, and '
-            'tiering will not evict any partitions. Enable CDF on the lakets_cdf '
+    -- wired before CDF is turned on. wal2delta is a Lakebase-managed schema; if
+    -- this role lacks USAGE on it, probing raises insufficient_privilege -- treat
+    -- that the same as "not visible" so enable_sync (which does not otherwise
+    -- touch wal2delta) still completes.
+    BEGIN
+        v_cdf_present := to_regclass('wal2delta.tables') IS NOT NULL;
+    EXCEPTION WHEN insufficient_privilege THEN
+        v_cdf_present := FALSE;
+    END;
+    IF NOT v_cdf_present THEN
+        RAISE WARNING 'Lakebase CDF (wal2delta) is not enabled or not visible to this '
+            'role. The shadow will be created but will not replicate to Unity Catalog, '
+            'and tiering will not evict any partitions. Enable CDF on the lakets_cdf '
             'schema (see the LakeTS prerequisites) to activate sync.';
     END IF;
 
diff --git a/website/docs/troubleshooting.md b/website/docs/troubleshooting.md
index 587028d..40bbd0d 100644
--- a/website/docs/troubleshooting.md
+++ b/website/docs/troubleshooting.md
@@ -93,6 +93,19 @@ Lakebase CDF is schema-sensitive. After an `ALTER TABLE`:
 
 You enabled sync on the ChronoTable itself instead of the shadow table. Make sure `_chronotable_registry.sync_enabled = true` and the registry's `shadow_table_name` is what's configured in the Databricks sync — not the partitioned parent.
 
+### `ERROR: permission denied for schema wal2delta`
+
+`wal2delta` is the Lakebase CDF subsystem's own schema. `enable_sync` and the tiering durability gate probe it only to check whether CDF is active; they never write to it. If your role lacks `USAGE` on `wal2delta`, the probe is tolerated — `enable_sync` completes (creating the shadow and trigger as usual) and the tiering gate fails closed (it evicts nothing it cannot verify).
+
+If you need CDF-gated tiering to actually evict partitions, grant the role read access to the gate it depends on:
+
+```sql
+GRANT USAGE ON SCHEMA wal2delta TO "<your-role>";
+GRANT SELECT ON wal2delta.tables TO "<your-role>";
+```
+
+Run these as the `wal2delta` owner or a Lakebase admin. Until then, tiering stays fail-closed by design.
+
 ## Performance
 
 ### Inserts slow down over time