diff --git a/vegafusion-core/src/planning/stitch.rs b/vegafusion-core/src/planning/stitch.rs
index ded556c6..6219eb21 100644
--- a/vegafusion-core/src/planning/stitch.rs
+++ b/vegafusion-core/src/planning/stitch.rs
@@ -6,6 +6,7 @@ use crate::spec::signal::SignalSpec;
 use crate::spec::values::MissingNullOrValue;
 use crate::task_graph::graph::ScopedVariable;
 use crate::task_graph::scope::TaskScope;
+use itertools::Itertools;
 use serde_json::Value;
 use std::collections::HashSet;
 
@@ -76,8 +77,8 @@ pub fn stitch_specs(
 
     // Return plan which specifies which signals/data need to be communicated between client and server
     Ok(CommPlan {
-        server_to_client: server_to_client.into_iter().collect(),
-        client_to_server: client_to_server.into_iter().collect(),
+        server_to_client: server_to_client.into_iter().sorted().collect(),
+        client_to_server: client_to_server.into_iter().sorted().collect(),
     })
 }
 
diff --git a/vegafusion-python/tests/test_pretransform.py b/vegafusion-python/tests/test_pretransform.py
index da8072fd..df326685 100644
--- a/vegafusion-python/tests/test_pretransform.py
+++ b/vegafusion-python/tests/test_pretransform.py
@@ -1761,9 +1761,9 @@ def test_pre_transform_spec_encoded_datasets():
     # Pre-transform with supported aggregate function should result in no warnings
     vega_spec = movies_histogram_spec()
 
-    # default list of dict format
-    tx_spec, _warnings = vf.runtime.pre_transform_spec(
-        vega_spec, data_encoding_threshold=10, data_encoding_format="pyarrow"
+    # Inline when threshold is larger than transformed data
+    tx_spec, datasets, _warnings = vf.runtime.pre_transform_extract(
+        vega_spec, extract_threshold=10, extracted_format="pyarrow"
     )
 
     values = tx_spec["data"][0]["values"]
@@ -1771,33 +1771,33 @@ def test_pre_transform_spec_encoded_datasets():
     assert len(values) == 9
 
     # pyarrow format
-    tx_spec, _warnings = vf.runtime.pre_transform_spec(
-        vega_spec, data_encoding_threshold=0, data_encoding_format="pyarrow"
+    tx_spec, datasets, _warnings = vf.runtime.pre_transform_extract(
+        vega_spec, extract_threshold=0, extracted_format="pyarrow"
     )
 
-    values = tx_spec["data"][0]["values"]
+    name, scope, values = datasets[0]
+    assert name == "source_0"
     assert isinstance(values, pa.Table)
     values_df = values.to_pandas()
     assert len(values_df) == 9
     assert values_df.columns[0] == "bin_maxbins_10_IMDB Rating"
 
     # arrow-ipc format
-    tx_spec, _warnings = vf.runtime.pre_transform_spec(
-        vega_spec, data_encoding_threshold=0, data_encoding_format="arrow-ipc"
+    tx_spec, datasets, _warnings = vf.runtime.pre_transform_extract(
+        vega_spec, extract_threshold=0, extracted_format="arrow-ipc"
     )
-
-    values = tx_spec["data"][0]["values"]
+    name, scope, values = datasets[0]
     assert isinstance(values, bytes)
     values_df = pa.ipc.deserialize_pandas(values)
     assert len(values_df) == 9
     assert values_df.columns[0] == "bin_maxbins_10_IMDB Rating"
 
     # arrow-ipc-base64 format
-    tx_spec, _warnings = vf.runtime.pre_transform_spec(
-        vega_spec, data_encoding_threshold=0, data_encoding_format="arrow-ipc-base64"
+    tx_spec, datasets, _warnings = vf.runtime.pre_transform_extract(
+        vega_spec, extract_threshold=0, extracted_format="arrow-ipc-base64"
     )
 
-    values = tx_spec["data"][0]["values"]
+    name, scope, values = datasets[0]
     assert isinstance(values, str)
     values_df = pa.ipc.deserialize_pandas(base64.standard_b64decode(values))
     assert len(values_df) == 9