diff --git a/CHANGELOG.md b/CHANGELOG.md
index 135f8422..bb502cb6 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -26,6 +26,8 @@
 
 * `run_benchmark`: Emit one dataset metadata entry per dataset by de-duplicating on `dataset_id`, rather than by keeping only the `log_cp10k` states. The old filter emitted nothing at all if a dataset ever arrived under a different normalization (PR #48).
 
+* `file_test_mod2`: Declare `uns["normalization_id"]`, which `run_benchmark` reads off this file to decide which method to run on which dataset (PR #30).
+
 ## BUG FIXES
 
 * `process_dataset`: Fall back to holding out a quarter of the batches when the dataset has no `obs["is_train"]`, rather than silently producing four empty h5ads. `obs["is_train"]` carries the NeurIPS 2021 competition split and stays optional; `obs["cell_type"]` is now declared and required (PR #28).
diff --git a/src/api/file_test_mod2.yaml b/src/api/file_test_mod2.yaml
index dc6417da..6110c1c6 100644
--- a/src/api/file_test_mod2.yaml
+++ b/src/api/file_test_mod2.yaml
@@ -80,6 +80,10 @@ info:
         type: string
         description: The organism of the sample in the dataset.
         required: false
+      - name: normalization_id
+        type: string
+        description: The unique identifier of the normalization method used.
+        required: true
       - type: string
         name: gene_activity_var_names
         description: "Names of the gene activity matrix"