diff --git a/orte/mca/gpr/replica/communications/gpr_replica_remote_msg.c b/orte/mca/gpr/replica/communications/gpr_replica_remote_msg.c
index 28b00e82e7..4d81c8e3d4 100644
--- a/orte/mca/gpr/replica/communications/gpr_replica_remote_msg.c
+++ b/orte/mca/gpr/replica/communications/gpr_replica_remote_msg.c
@@ -80,9 +80,27 @@ int orte_gpr_replica_remote_notify(orte_process_name_t *recipient,
     
     if (0 > orte_rml.send_buffer_nb(recipient, buffer, ORTE_RML_TAG_GPR_NOTIFY, 0,
                                     orte_gpr_replica_remote_send_cb, NULL)) {
+#if 0
+        /* temporarily disable this error report
+         * With the new orted-failed-to-start code, we hold a caller in
+         * the rmgr.spawn function until either the app launches or
+         * it fails. Failure is indicated by a subscription to NUM_TERMINATED.
+         * However, that means that a notify_msg is going to get sent to a
+         * remote process during comm_spawn once all procs terminate. Since
+         * that process will have terminated, and the HNP processes the trigger
+         * first, the notify_msg send will fail as the recipient will have
+         * terminated and exited.
+         *
+         * A proper fix will require that we do something different
+         * in rmgr_proxy.spawn so we don't get a callback after the
+         * process is done
+         */
         ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+        opal_output(0, "send failed to [%ld,%ld,%ld]", ORTE_NAME_ARGS(recipient));
+        orte_dss.dump(0, message, ORTE_GPR_NOTIFY_MSG);
         OPAL_THREAD_LOCK(&orte_gpr_replica_globals.mutex);
         return ORTE_ERR_COMM_FAILURE;
+#endif
     }
     OPAL_THREAD_LOCK(&orte_gpr_replica_globals.mutex);