diff --git a/src/BioSimSpace/FreeEnergy/_relative.py b/src/BioSimSpace/FreeEnergy/_relative.py
index 98801998..19bc9e08 100644
--- a/src/BioSimSpace/FreeEnergy/_relative.py
+++ b/src/BioSimSpace/FreeEnergy/_relative.py
@@ -1059,12 +1059,16 @@ def _somd2_extract(parquet_file, T=None, estimator="MBAR"):
             raise ValueError("Parquet metadata does not contain 'lambda'.")
         if not is_mbar:
             try:
-                lambda_grad = metadata["lambda_grad"]
+                # Normalise to floats to match the DataFrame column type expected
+                # by alchemlyb (handles both old float and new string metadata).
+                lambda_grad = [float(v) for v in metadata["lambda_grad"]]
             except:
                 raise ValueError("Parquet metadata does not contain 'lambda grad'")
         else:
             try:
-                lambda_grad = metadata["lambda_grad"]
+                # Normalise to floats to match the DataFrame column type expected
+                # by alchemlyb (handles both old float and new string metadata).
+                lambda_grad = [float(v) for v in metadata["lambda_grad"]]
             except:
                 lambda_grad = []
 
@@ -1078,6 +1082,19 @@ def _somd2_extract(parquet_file, T=None, estimator="MBAR"):
         # Convert to a pandas dataframe.
         df = table.to_pandas()
 
+        # Normalise column names to floats so that comparisons are consistent
+        # regardless of whether the parquet was written with float keys (old
+        # sire) or formatted string keys (new sire). float("0.10000") and
+        # float("0.1") give the same IEEE754 value, so old and new files are
+        # handled identically and the alchemlyb index check passes.
+        df.columns = [
+            float(c)
+            if isinstance(c, str)
+            and c.replace(".", "", 1).replace("-", "", 1).isdigit()
+            else c
+            for c in df.columns
+        ]
+
         if is_mbar:
             # Extract all columns other than those used for the gradient.
             df = df[[x for x in df.columns if x not in lambda_grad]]