diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 4a45a0d..f56a39f 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -119,5 +119,4 @@ jobs:
         run: |
           pytest Tests/eval -v \
             --ignore=Tests/eval/test_joint_embedding.py \
-            --deselect Tests/eval/test_eval_stats.py::test_mann_whitney_u_disjoint \
             --deselect Tests/eval/test_embedding_space.py::test_cka_independent
diff --git a/Tests/eval/test_eval_stats.py b/Tests/eval/test_eval_stats.py
index ba00050..ac3d6a7 100644
--- a/Tests/eval/test_eval_stats.py
+++ b/Tests/eval/test_eval_stats.py
@@ -42,9 +42,15 @@ def test_cohens_d_large_nondegenerate_effect():
 
 
 def test_mann_whitney_u_disjoint():
-    result = mann_whitney_u([1, 2, 3], [10, 11, 12])
+    # Four per group, not three: with n1 = n2 = 3 the smallest attainable
+    # two-sided p is 2/20 = 0.1, so no correct implementation can clear 0.05
+    # there however cleanly the groups separate. At four per group the exact
+    # minimum is 2/70 = 0.029.
+    result = mann_whitney_u([1, 2, 3, 4], [10, 11, 12, 13])
+    assert result["u_statistic"] == 0.0
     assert result["p_value"] < 0.05
-    assert "u_statistic" in result
+    assert result["n_a"] == 4
+    assert result["n_b"] == 4
 
 
 def test_bonferroni_correct():