Bugfix get_untrustworthy_fields() for score outputs

CHANGELOG.md

@@ -16,7 +16,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [0.0.2] 2026-02-22
 - Improve structured outputs functionality + tutorial
 
-[Unreleased]: https://github.com/cleanlab/tlm/compare/v0.0.2...HEAD
+## [0.0.3] 2026-02-22
+- Fix `get_untrustworthy_fields()` for score outputs
+
+[Unreleased]: https://github.com/cleanlab/tlm/compare/v0.0.3...HEAD
+[0.0.3]: https://github.com/cleanlab/tlm/commits/v0.0.3
 [0.0.2]: https://github.com/cleanlab/tlm/commits/v0.0.2
 [0.0.1]: https://github.com/cleanlab/tlm/commits/v0.0.1
 [0.0.0]: https://github.com/cleanlab/tlm/commits/v0.0.0

tlm/__about__.py

@@ -1 +1 @@
-__version__ = "0.0.2"
+__version__ = "0.0.3"

tlm/utils/structured_output_utils.py

@@ -10,24 +10,36 @@ def _get_untrustworthy_fields(
     display_details: bool = True,
 ) -> list[str]:
     tlm_metadata = tlm_result["metadata"]
-    response_text = tlm_result["response"].choices[0].message.content  # type: ignore
+    response = tlm_result["response"]
+
+    # for score completions
+    if isinstance(response, dict) and "chat_completion" in response:
+        response = response["chat_completion"]
+
+    try:
+        response_text = response.choices[0].message.content  # type: ignore
+    except Exception:
+        # sometimes tlm_result["response"] is a dictionary
+        response_text = response["choices"][0]["message"]["content"]  # type: ignore
 
     if tlm_metadata is None or "per_field_score" not in tlm_metadata:
         raise ValueError(
             "`per_field_score` is not present in the metadata.\n"
             "`get_untrustworthy_fields()` can only be called scoring structured outputs responses."
         )
 
-    try:
-        so_response = json.loads(response_text)
-    except Exception:
-        pass
-    try:
-        so_response = ast.literal_eval(response_text)
-    except Exception:
-        raise ValueError(
-            "The LLM response must be a valid JSON output (use `response_format` to specify the output format)"
-        )
+    if isinstance(response_text, dict):
+        so_response = response_text
+    else:
+        try:
+            so_response = json.loads(response_text)
+        except Exception:
+            try:
+                so_response = ast.literal_eval(response_text)
+            except Exception:
+                raise ValueError(
+                    "The LLM response must be a valid JSON output (use `response_format` to specify the output format)"
+                )
 
     per_field_score = tlm_metadata["per_field_score"]
     per_score_details = []