test: use eligible benchmarks in harness ranking fixture

tests/test_ranking_refinements.py

@@ -107,12 +107,10 @@ def test_ranking_plot_height_scales_with_agents_and_keeps_all_y_labels():
 def test_pairwise_harness_rankings_exclude_low_data_harnesses(tmp_path):
     rankings = _rankings_module()
     csv_path = tmp_path / "results.csv"
+    # Use eligible benchmarks so this exercises harness exclusions, not the benchmark allowlist.
     pd.DataFrame(
         {
-            "benchmark.name": [
-                "Bench A", "Bench A", "Bench A", "Bench A",
-                "Bench B", "Bench B", "Bench B", "Bench B",
-            ],
+            "benchmark.name": ["RH SWE-Bench"] * 4 + ["SWE-Bench Verified"] * 4,
             "model.name": ["m1", "m1", "m1", "m1", "m2", "m2", "m2", "m2"],
             "harness.name": [
                 "Claude Code", "OpenCode", "Codex", "Qwen Code",
@@ -136,7 +134,7 @@ def test_pairwise_harness_rankings_exclude_low_data_harnesses(tmp_path):
     assert set(model_df["Category"]) == {"m1", "m2"}
 
     model_by_category = model_df.set_index("Category")
-    assert model_by_category.loc["m1", "Bench A"] == pytest.approx(0.838)
+    assert model_by_category.loc["m1", "RH SWE-Bench"] == pytest.approx(0.838)
 
 
 def test_ranking_minimum_height_is_reduced_without_changing_matrix_minimum():