diff --git a/src/oumi/analyze/utils/dataframe.py b/src/oumi/analyze/utils/dataframe.py
index 26ce8f0c18..66cac3b987 100644
--- a/src/oumi/analyze/utils/dataframe.py
+++ b/src/oumi/analyze/utils/dataframe.py
@@ -28,7 +28,10 @@
 
 def to_analysis_dataframe(
     conversations: list[Conversation],
-    results: Mapping[str, Sequence[BaseModel] | BaseModel],
+    results: Mapping[
+        str,
+        Sequence[BaseModel | dict[str, Any]] | BaseModel | dict[str, Any],
+    ],
     message_to_conversation_idx: list[int] | None = None,
 ) -> pd.DataFrame:
     """Convert typed analysis results to a pandas DataFrame.
@@ -117,7 +120,9 @@ def to_analysis_dataframe(
                             "missing values.",
                         )
 
-            elif isinstance(analyzer_results, BaseModel):
+            elif isinstance(analyzer_results, (BaseModel, dict)):
+                # Dataset-level result — same for every conversation. May be a
+                # raw dict when loaded from the pipeline's JSON cache.
                 _add_result_to_row(row, analyzer_results, prefix)
 
         rows.append(row)
@@ -211,7 +216,10 @@ def _add_result_to_row(
 
 
 def results_to_dict(
-    results: Mapping[str, Sequence[BaseModel] | BaseModel],
+    results: Mapping[
+        str,
+        Sequence[BaseModel | dict[str, Any]] | BaseModel | dict[str, Any],
+    ],
 ) -> dict[str, list[dict[str, Any]] | dict[str, Any]]:
     """Convert typed results to a serializable dictionary.
 
@@ -227,8 +235,12 @@ def results_to_dict(
 
     for name, result in results.items():
         if isinstance(result, list):
-            output[name] = [r.model_dump() for r in result]
+            output[name] = [
+                r.model_dump() if isinstance(r, BaseModel) else r for r in result
+            ]
         elif isinstance(result, BaseModel):
             output[name] = result.model_dump()
+        elif isinstance(result, dict):
+            output[name] = result
 
     return output
diff --git a/tests/unit/analyze/test_dataframe_utils.py b/tests/unit/analyze/test_dataframe_utils.py
index 3263a4d1d5..c05dbd0299 100644
--- a/tests/unit/analyze/test_dataframe_utils.py
+++ b/tests/unit/analyze/test_dataframe_utils.py
@@ -417,3 +417,52 @@ def test_results_to_dict_empty():
     output = results_to_dict({})
 
     assert output == {}
+
+
+# -----------------------------------------------------------------------------
+# Tests: Cached (raw dict) results
+# -----------------------------------------------------------------------------
+
+
+def test_dataset_level_cached_dict_in_dataframe(sample_conversations):
+    """Dataset-level results as raw dicts (from cache) are included."""
+    results = {
+        "Stats": {"total_count": 100, "avg_score": 75.5},
+    }
+
+    df = to_analysis_dataframe(sample_conversations, results)
+
+    assert "stats__total_count" in df.columns
+    assert "stats__avg_score" in df.columns
+    assert (df["stats__total_count"] == 100).all()
+    assert (df["stats__avg_score"] == 75.5).all()
+
+
+def test_per_conversation_cached_dicts_in_dataframe(sample_conversations):
+    """Per-conversation results as raw dicts (from cache) work."""
+    results = {
+        "Simple": [
+            {"score": 80, "name": "a"},
+            {"score": 90, "name": "b"},
+        ],
+    }
+
+    df = to_analysis_dataframe(sample_conversations, results)
+
+    assert "simple__score" in df.columns
+    assert "simple__name" in df.columns
+    assert df["simple__score"].tolist() == [80, 90]
+    assert df["simple__name"].tolist() == ["a", "b"]
+
+
+def test_results_to_dict_passes_through_cached_dicts():
+    """Raw dict results pass through without model_dump."""
+    results = {
+        "Stats": {"total_count": 100},
+        "PerConv": [{"score": 1}, {"score": 2}],
+    }
+
+    output = results_to_dict(results)
+
+    assert output["Stats"] == {"total_count": 100}
+    assert output["PerConv"] == [{"score": 1}, {"score": 2}]