fix: validate weight keys against selected clustering input, not full dataset

- Validate against ds_for_clustering (after data_vars selection) instead
  of ds (full dataset), so weights for excluded variables are caught
- Move validation before drop_constant_arrays so constant columns remain
  accepted as valid keys
- Fix test to use to_dataset() instead of clustering_data() so the
  constant-column path is actually exercised

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: FBumann

flixopt/transform_accessor.py

@@ -1694,20 +1694,20 @@ def cluster(
         else:
             ds_for_clustering = ds
 
-        # Filter constant arrays once on the full dataset (not per slice)
-        # This ensures all slices have the same variables for consistent metrics
-        ds_for_clustering = drop_constant_arrays(ds_for_clustering, dim='time')
-
-        # Validate user-provided weight keys against available variables
+        # Validate user-provided weight keys against the selected clustering input
         if cluster is not None and cluster.weights is not None:
-            all_vars = set(ds.data_vars)
-            unknown = sorted(set(cluster.weights) - all_vars)
+            selected_vars = set(ds_for_clustering.data_vars)
+            unknown = sorted(set(cluster.weights) - selected_vars)
             if unknown:
                 raise ValueError(
                     f'ClusterConfig weights reference unknown variables: {unknown}. '
                     f'Available variables can be found via transform.clustering_data().'
                 )
 
+        # Filter constant arrays once on the full dataset (not per slice)
+        # This ensures all slices have the same variables for consistent metrics
+        ds_for_clustering = drop_constant_arrays(ds_for_clustering, dim='time')
+
         # Guard against empty dataset after removing constant arrays
         if not ds_for_clustering.data_vars:
             filter_info = f'data_vars={data_vars}' if data_vars else 'all variables'

tests/test_clustering/test_integration.py

@@ -396,9 +396,12 @@ def test_extra_weight_keys_filtered_with_constant_column(self):
         constant_sink = Sink('constant_load', inputs=[constant_flow])
         fs.add_elements(source, sink, constant_sink, bus)
 
-        # The constant column name (find it from clustering_data)
-        all_data = fs.transform.clustering_data()
+        # Use to_dataset() to get ALL columns including the constant one
+        # (clustering_data() already strips constants, so it wouldn't test the path)
+        all_data = fs.to_dataset(include_solution=False)
         all_columns = set(all_data.data_vars)
+        clustering_columns = set(fs.transform.clustering_data().data_vars)
+        assert all_columns > clustering_columns, 'Test requires at least one constant column to be meaningful'
 
         # Build weights that reference ALL columns including the constant one
         # that will be dropped — these are valid variables, just constant over time