FIX: Match get_scores score_category against each stored category

pyrit/memory/memory_interface.py

@@ -3502,7 +3502,7 @@ def _execute_get_scores(
         *,
         score_ids: Sequence[str] | None = None,
         score_type: str | None = None,
-        score_category: str | None = None,
+        score_category: str | Sequence[str] | None = None,
         sent_after: datetime | None = None,
         sent_before: datetime | None = None,
         identifier_filters: Sequence[IdentifierFilter] | None = None,
@@ -3514,7 +3514,8 @@ def _execute_get_scores(
         Args:
             score_ids (Sequence[str] | None): A list of score IDs to filter by.
             score_type (str | None): The type of the score to filter by.
-            score_category (str | None): The category of the score to filter by.
+            score_category (str | Sequence[str] | None): Return scores that list this category, or every category
+                of a sequence. Names match whole and case-insensitively.
             sent_after (datetime | None): Filter for scores sent after this datetime.
             sent_before (datetime | None): Filter for scores sent before this datetime.
             identifier_filters (Sequence[IdentifierFilter] | None): A sequence of IdentifierFilter objects that
@@ -3533,7 +3534,11 @@ def _execute_get_scores(
         if score_type:
             conditions.append(ScoreEntry.score_type == score_type)
         if score_category:
-            conditions.append(ScoreEntry.score_category == score_category)
+            # score_category is stored as a JSON list, so match one whole element of it.
+            categories = [score_category] if isinstance(score_category, str) else list(score_category)
+            self._add_list_conditions(
+                field=ScoreEntry.score_category, conditions=conditions, values=categories, exact=True
+            )
         if sent_after:
             conditions.append(ScoreEntry.timestamp >= sent_after)
         if sent_before:
@@ -9692,7 +9697,7 @@ def get_scores(
         *,
         score_ids: Sequence[str] | None = None,
         score_type: str | None = None,
-        score_category: str | None = None,
+        score_category: str | Sequence[str] | None = None,
         sent_after: datetime | None = None,
         sent_before: datetime | None = None,
         identifier_filters: Sequence[IdentifierFilter] | None = None,
@@ -9725,7 +9730,7 @@ async def get_scores_async(
         *,
         score_ids: Sequence[str] | None = None,
         score_type: str | None = None,
-        score_category: str | None = None,
+        score_category: str | Sequence[str] | None = None,
         sent_after: datetime | None = None,
         sent_before: datetime | None = None,
         identifier_filters: Sequence[IdentifierFilter] | None = None,
@@ -9737,7 +9742,8 @@ async def get_scores_async(
         Args:
             score_ids (Sequence[str] | None): A list of score IDs to filter by.
             score_type (str | None): The type of the score to filter by.
-            score_category (str | None): The category of the score to filter by.
+            score_category (str | Sequence[str] | None): Return scores that list this category, or every category
+                of a sequence. Names match whole and case-insensitively.
             sent_after (datetime | None): Filter for scores sent after this datetime.
             sent_before (datetime | None): Filter for scores sent before this datetime.
             identifier_filters (Sequence[IdentifierFilter] | None): A sequence of IdentifierFilter objects that

tests/unit/memory/memory_interface/test_interface_scores.py

@@ -448,3 +448,47 @@ async def test_get_scores_by_scorer_identifier_filter(
         ],
     )
     assert len(results) == 0
+
+
+async def test_get_scores_by_score_category(
+    sqlite_instance: MemoryInterface,
+    sample_conversation_entries: Sequence[PromptMemoryEntry],
+):
+    prompt_id = sample_conversation_entries[0].id
+    sqlite_instance._insert_entries(entries=sample_conversation_entries)
+
+    def make_score(value: str, categories: list[str] | None, score_type: str = "float_scale") -> Score:
+        return Score(
+            score_value=value,
+            score_value_description="",
+            score_type=score_type,
+            score_category=categories,
+            score_rationale="",
+            score_metadata={},
+            scorer_class_identifier=_test_scorer_id(),
+            message_piece_id=prompt_id,
+        )
+
+    await sqlite_instance.add_scores_to_memory_async(
+        scores=[
+            make_score("0.1", ["hate"]),
+            make_score("0.2", ["violence", "hate"]),
+            make_score("0.3", ["hate_speech"]),
+            make_score("0.4", ["HaTe"]),
+            make_score("0.5", None),
+            make_score("0.6", []),
+            make_score("true", ["hate"], score_type="true_false"),
+        ]
+    )
+
+    async def values_for(category: str | list[str], score_type: str | None = None):
+        scores = await sqlite_instance.get_scores_async(score_category=category, score_type=score_type)
+        return sorted(score.score_value for score in scores)
+
+    assert await values_for("hate") == ["0.1", "0.2", "0.4", "true"]
+    assert await values_for("HATE", score_type="float_scale") == ["0.1", "0.2", "0.4"]
+    assert await values_for("violence") == ["0.2"]
+    assert await values_for("hate_speech") == ["0.3"]
+    assert await values_for("speech") == []
+    assert await values_for(["hate", "violence"]) == ["0.2"]
+    assert await values_for(["hate"], score_type="float_scale") == ["0.1", "0.2", "0.4"]