diff --git a/mdx/int_edugain/verbs.xml b/mdx/int_edugain/verbs.xml
index a540a5d5..015037a1 100644
--- a/mdx/int_edugain/verbs.xml
+++ b/mdx/int_edugain/verbs.xml
@@ -27,6 +27,11 @@
     -->
     <import resource="classpath:uk/org/iay/incommon/mda/beans.xml"/>
 
+    <!--
+        Import ukf channel beans, for uk_assemble.
+    -->
+    <import resource="classpath:uk/beans.xml"/>
+
     <!--
         Import UK federation ingress policy for eduGAIN.
     -->
@@ -179,8 +184,24 @@
                 <ref bean="int_edugain_productionEntities"/>
                 <ref bean="removeUKEntities"/>
 
+                <!--
+                    Sort the remaining entities by entity ID, so that any errors
+                    are reported in a stable order, even if some registrars
+                    have unstable output.
+                    
+                    Ideally this should be handled by an explicit ordering stage.
+                    Lacking one, we use the fact that we can use the ordering
+                    strategy ability of the EntitiesDescriptor assembler to
+                    order entities, and then disassemble again. Ordering is based
+                    on the ItemID populated into the item metadata.
+                -->
+                <ref bean="populateItemIds"/>
+                <ref bean="uk_assemble"/>
+                <ref bean="disassemble"/>
+
                 <ref bean="standardImportActions"/>
                 <ref bean="edugainPolicy"/>
+
                 <ref bean="warningAndErrorAnnouncer"/>
                 <ref bean="errorTerminator"/>
             </list>
diff --git a/mdx/uk/verbs.xml b/mdx/uk/verbs.xml
index 22eb99a0..58c3ce62 100644
--- a/mdx/uk/verbs.xml
+++ b/mdx/uk/verbs.xml
@@ -146,11 +146,15 @@
                 <ref bean="uk_registeredEntities"/>
 
                 <!--
-                    The next four stages ensure that any errors are reported
+                    The next five stages ensure that any errors are reported
                     in entity ID order, even on systems where the native order
                     (which tends to reflect the file enumeration ordering)
-                    is arbitrary.  Ideally this should be handled by an optional
-                    ordering strategy on the announcing filter.
+                    is arbitrary.  Ideally this should be handled by an explicit
+                    ordering stage.
+
+                    The actual reordering is performed by the uk_assemble/disassemble
+                    stages; the subsequent "populate" stages restore the item
+                    metadata discarded by the uk_assemble stage.
                 -->
                 <ref bean="uk_assemble"/>
                 <ref bean="disassemble"/>
@@ -178,11 +182,15 @@
                 <ref bean="uk_registeredEntities"/>
 
                 <!--
-                    The next four stages ensure that any errors are reported
+                    The next five stages ensure that any errors are reported
                     in entity ID order, even on systems where the native order
                     (which tends to reflect the file enumeration ordering)
-                    is arbitrary.  Ideally this should be handled by an optional
-                    ordering strategy on the announcing filter.
+                    is arbitrary.  Ideally this should be handled by an explicit
+                    ordering stage.
+
+                    The actual reordering is performed by the uk_assemble/disassemble
+                    stages; the subsequent "populate" stages restore the item
+                    metadata discarded by the uk_assemble stage.
                 -->
                 <ref bean="uk_assemble"/>
                 <ref bean="disassemble"/>