fix: update packet evaluation metric

src/stickler/doc_split/packet_evaluation_metrics.py

@@ -134,7 +134,7 @@ def calculate_ordering_score_per_group(
     """
     Calculate Kendall's Tau for each document group.
 
-    Single-page groups are excluded (ordering undefined).
+    Single-page groups are assigned a perfect score of 1.0 (trivially in correct order).
 
     Args:
         data: DataFrame with group_id, page_number, page_number_predicted.
@@ -151,6 +151,7 @@ def calculate_ordering_score_per_group(
 
     for group_id, group_data in data.groupby("group_id"):
         if len(group_data) <= 1:
+            group_scores[group_id] = 1.0
             continue
 
         tau, _p_value = kendalltau(
@@ -163,9 +164,9 @@ def calculate_ordering_score_per_group(
 
 def calculate_average_ordering_score(group_scores: Dict[Any, float]) -> float:
     """
-    Mean Kendall's Tau across all multi-page groups.
+    Mean Kendall's Tau across all groups (single-page groups score 1.0).
 
-    Returns 0 if no multi-page groups exist.
+    Returns 0 if no groups exist.
     """
     if not group_scores:
         return 0

tests/doc_split/test_packet_evaluation_metrics.py

@@ -308,16 +308,17 @@ def test_reverse_ordering(self):
         avg = calculate_average_ordering_score(scores)
         assert avg == pytest.approx(-1.0, abs=1e-4)
 
-    def test_single_page_groups_excluded(self):
-        """Groups with only 1 page should not appear in ordering scores."""
+    def test_single_page_groups_score_perfect(self):
+        """Single-page groups should receive a perfect ordering score of 1.0."""
         data = [
             _page("invoice", "inv-01", 1, "invoice", "inv-01", 1),
             _page("form", "form-01", 2, "form", "form-01", 2),
         ]
         df = pd.DataFrame(data)
         scores = calculate_ordering_score_per_group(df)
-        assert len(scores) == 0
-        assert calculate_average_ordering_score(scores) == 0
+        assert len(scores) == 2
+        assert all(v == 1.0 for v in scores.values())
+        assert calculate_average_ordering_score(scores) == 1.0
 
     def test_missing_columns_raises(self):
         df = pd.DataFrame([{"foo": 1}])