[AIC-py][eval] make e2e test a little more robust (#435)

[AIC-py][eval] make e2e test a little more robust

---
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with
[ReviewStack](https://reviewstack.dev/lastmile-ai/aiconfig/pull/435).
* __->__ #435
* #432

python/tests/test_eval.py

@@ -140,8 +140,6 @@ async def test_run_test_suite_with_inputs(data: st.DataObject):
         )
     )
 
-    input_data, _ = cu.unzip(user_test_suite_with_inputs)
-
     mock_aiconfig = MockAIConfigRuntime()
 
     out = await run_test_suite_helper(
@@ -166,13 +164,25 @@ async def test_run_test_suite_with_inputs(data: st.DataObject):
                 "best_possible_value",
                 "worst_possible_value",
             ]
-            inputs = df["input"].astype(str).tolist()  # type: ignore[no-untyped-call]
-            assert set(inputs) == set(input_data)  # type: ignore[no-untyped-call]
+
+            input_pairs = {
+                (input_datum, metric.interpretation.id)
+                for input_datum, metric in user_test_suite_with_inputs
+            }
+            result_pairs = set(  # type: ignore[no-untyped-call]
+                df[["input", "metric_id"]].itertuples(index=False, name=None)  # type: ignore[no-untyped-call]
+            )
+
+            assert input_pairs == result_pairs
 
             df_brevity = df[df["metric_name"] == "brevity"]
             assert (
                 df_brevity["aiconfig_output"].apply(len)  # type: ignore[no-untyped-call]
                 == df_brevity["value"]  # type: ignore[no-untyped-call]
             ).all()
+
+            df_substring = df[df["metric_name"] == "substring_match"]
+            assert (df_substring["value"].apply(lambda x: x in {0.0, 0.1})).all()  # type: ignore[no-untyped-call]
+
         case Err(e):
             assert False, f"expected Ok, got Err({e})"