Merge main and address ranking review feedback

src/charts.py

@@ -24,7 +24,7 @@ PlotBackground = Literal["Dark", "White"]
 DEFAULT_PALETTE: PaletteName = "Citrus"
 DEFAULT_BACKGROUND: PlotBackground = "Dark"
 
-RANKING_MIN_HEIGHT_PX = 680
+RANKING_MIN_HEIGHT_PX = 340
 RANKING_ROW_HEIGHT_PX = 44
 RANKING_VERTICAL_PADDING_PX = 280
 MATRIX_MIN_HEIGHT_PX = 720

src/rankings.py

@@ -7,6 +7,7 @@ import choix
 
 
 EXCLUDED_BENCHMARKS = {"Shellbench"}
+EXCLUDED_HARNESSES_FROM_PAIRWISE = {"Codex", "Qwen Code"}
 
 
 def prepare_ranking_data(
@@ -147,10 +148,18 @@ def load_and_rank(
         empty = _empty_table()
         return empty, empty, empty
 
+    pairwise_harness_df = df.loc[
+        ~df["harness.name"].isin(EXCLUDED_HARNESSES_FROM_PAIRWISE)
+    ].reset_index(drop=True)
+
     results = []
-    for rank_fn in (rank_harnesses, rank_models, rank_pairs):
+    for rank_fn, ranking_df in (
+        (rank_harnesses, pairwise_harness_df),
+        (rank_models, df),
+        (rank_pairs, pairwise_harness_df),
+    ):
         try:
-            results.append(rank_fn(df, metcol, descending))
+            results.append(rank_fn(ranking_df, metcol, descending))
         except ValueError:
             results.append(_empty_table())
     return results[0], results[1], results[2]

tests/test_ranking_refinements.py

@@ -90,3 +90,37 @@ def test_ranking_plot_height_scales_with_agents_and_keeps_all_y_labels():
     assert figure.layout.yaxis.tickmode == "array"
     assert len(figure.layout.yaxis.tickvals) == rows
     assert len(figure.layout.yaxis.ticktext) == rows
+
+
+def test_pairwise_harness_rankings_exclude_low_data_harnesses(tmp_path):
+    rankings = _rankings_module()
+    csv_path = tmp_path / "results.csv"
+    pd.DataFrame(
+        {
+            "benchmark.name": [
+                "Bench A", "Bench A", "Bench A", "Bench A",
+                "Bench B", "Bench B", "Bench B", "Bench B",
+            ],
+            "model.name": ["m1", "m1", "m1", "m1", "m2", "m2", "m2", "m2"],
+            "harness.name": [
+                "Claude Code", "OpenCode", "Codex", "Qwen Code",
+                "Claude Code", "OpenCode", "Codex", "Qwen Code",
+            ],
+            "metrics.score": [0.8, 0.7, 0.95, 0.9, 0.75, 0.65, 0.93, 0.88],
+            "model.is_oss": [True] * 8,
+            "harness.is_oss": [True] * 8,
+        }
+    ).to_csv(csv_path, index=False)
+
+    harness_df, model_df, pair_df = rankings.load_and_rank(csv_path)
+
+    assert set(harness_df["Category"]) == {"Claude Code", "OpenCode"}
+    assert not pair_df["Category"].str.contains("Codex|Qwen Code").any()
+    assert set(model_df["Category"]) == {"m1", "m2"}
+
+
+def test_ranking_minimum_height_is_reduced_without_changing_matrix_minimum():
+    source = Path("src/charts.py").read_text()
+
+    assert "RANKING_MIN_HEIGHT_PX = 340" in source
+    assert "MATRIX_MIN_HEIGHT_PX = 720" in source