implemented Marcel's suggestions

columnflow/production/normalization.py

@@ -176,7 +176,12 @@ def normalization_weights(self: Producer, events: ak.Array, **kwargs) -> ak.Arra
     attribute. When py:attr`allow_stitching` is set to True, the sum of event weights is computed
     for all datasets with a leaf process contained in the leaf processes of the
     py:attr:`dataset_inst`. For stitching, the process_id needs to be reconstructed for each leaf
-    process on a per event basis.
+    process on a per event basis. Moreover, when stitching is enabled, the an additional normalization
+    weight is computed for the inclusive dataset only and stored in a column named
+    py:attr:`weight_name`_inclusive_only. This weight resembles the normalization weight for the
+    inclusive dataset, as if it were unstitched and should therefore only be applied, when using the
+    inclusive dataset as a standalone dataset.
+
     """
     # read the process id column
     process_id = np.asarray(events.process_id)
@@ -198,9 +203,9 @@ def normalization_weights(self: Producer, events: ak.Array, **kwargs) -> ak.Arra
     events = set_ak_column(events, self.weight_name, norm_weight, value_type=np.float32)
 
     # If we are stitching, we also compute the inclusive weight for debugging purposes
-    if self.allow_stitching:
+    if self.allow_stitching and self.dataset_inst == self.get_inclusive_dataset():
         incl_norm_weight = events.mc_weight * self.inclusive_weight
-        events = set_ak_column(events, f"{self.weight_name}_incl", incl_norm_weight, value_type=np.float32)
+        events = set_ak_column(events, f"{self.weight_name}_inclusive_only", incl_norm_weight, value_type=np.float32)
     return events
 
 
@@ -210,11 +215,6 @@ def normalization_weights_requires(self: Producer, reqs: dict) -> None:
     Adds the requirements needed by the underlying py:attr:`task` to access selection stats into
     *reqs*.
     """
-    if self.allow_stitching:
-        self.stitching_datasets = self.get_stitching_datasets()
-    else:
-        self.stitching_datasets = [self.dataset_inst]
-
     # check that all datasets are known
     for dataset in self.stitching_datasets:
         if not self.config_inst.has_dataset(dataset):
@@ -341,7 +341,12 @@ def normalization_weights_init(self: Producer) -> None:
     """
     self.produces.add(self.weight_name)
     if self.allow_stitching:
-        self.produces.add(f"{self.weight_name}_incl")
+        self.stitching_datasets = self.get_stitching_datasets()
+    else:
+        self.stitching_datasets = [self.dataset_inst]
+
+    if self.allow_stitching and self.dataset_inst == self.get_inclusive_dataset():
+        self.produces.add(f"{self.weight_name}_inclusive_only")
 
 
 stitched_normalization_weights = normalization_weights.derive(