diff --git a/test/train-sets/ref/cbe_adf_cover.stderr b/test/train-sets/ref/cbe_adf_cover.stderr
index 2c37e2d40ae..06fc981e901 100644
--- a/test/train-sets/ref/cbe_adf_cover.stderr
+++ b/test/train-sets/ref/cbe_adf_cover.stderr
@@ -1,4 +1,5 @@
 predictions = cbe_adf_cover.predict
+warning: currently, mtr is only used for the first policy in cover, other policies use dr
 Num weight bits = 18
 learning rate = 0.5
 initial_t = 0