Add warning for name collisions and test for include_local

- Warn when local dataset shadows a committed dataset with same name
- Add test for include_local=True behavior
- Add test for name collision warning

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <[email protected]>

Author: jucor

delphi/polismath/regression/datasets.py

@@ -89,7 +89,17 @@ def discover_datasets(include_local: bool = False) -> Dict[str, DatasetInfo]:
     """Auto-discover datasets from real_data/ and optionally .local/"""
     datasets = _discover_in_dir(get_real_data_dir(), is_local=False)
     if include_local:
-        datasets.update(_discover_in_dir(get_local_data_dir(), is_local=True))
+        local_datasets = _discover_in_dir(get_local_data_dir(), is_local=True)
+        # Warn about name collisions (local would shadow committed)
+        collisions = set(datasets.keys()) & set(local_datasets.keys())
+        if collisions:
+            import warnings
+            warnings.warn(
+                f"Local datasets shadow committed datasets with same name: {', '.join(sorted(collisions))}. "
+                f"Local versions will be used.",
+                UserWarning
+            )
+        datasets.update(local_datasets)
     return datasets
 
 

delphi/tests/test_datasets.py

@@ -88,6 +88,16 @@ def test_discover_real_data(self):
         result = discover_datasets(include_local=False)
         assert len(result) > 0
 
+    def test_discover_with_include_local(self):
+        """Should include local datasets when flag is set."""
+        without_local = discover_datasets(include_local=False)
+        with_local = discover_datasets(include_local=True)
+        # With local should have >= datasets (may have more if .local/ has data)
+        assert len(with_local) >= len(without_local)
+        # All committed datasets should still be present
+        for name in without_local:
+            assert name in with_local
+
     def test_list_regression_datasets(self):
         result = list_regression_datasets()
         assert isinstance(result, list)
@@ -101,3 +111,36 @@ def test_list_available_datasets(self):
     def test_paths_exist(self):
         assert get_real_data_dir().exists()
         assert get_local_data_dir().parent.exists()
+
+
+class TestNameCollision:
+    def test_warns_on_name_collision(self, tmp_path, monkeypatch):
+        """Should warn when local dataset shadows committed dataset."""
+        import warnings
+        from polismath.regression import datasets
+
+        # Create mock directories
+        real_data = tmp_path / "real_data"
+        local_data = real_data / ".local"
+        real_data.mkdir()
+        local_data.mkdir()
+
+        # Create dataset with same name in both locations
+        (real_data / "rabc123-test").mkdir()
+        (local_data / "rdef456-test").mkdir()  # Same name "test", different report_id
+
+        # Patch the directory functions
+        monkeypatch.setattr(datasets, "get_real_data_dir", lambda: real_data)
+        monkeypatch.setattr(datasets, "get_local_data_dir", lambda: local_data)
+
+        # Should warn about collision
+        with warnings.catch_warnings(record=True) as w:
+            warnings.simplefilter("always")
+            result = datasets.discover_datasets(include_local=True)
+            assert len(w) == 1
+            assert "shadow" in str(w[0].message)
+            assert "test" in str(w[0].message)
+
+        # Local version should win
+        assert result["test"].report_id == "rdef456"
+        assert result["test"].is_local