FIX: map canonical harm categories in get_all_harm_metrics

pyrit/score/scorer_evaluation/scorer_metrics_io.py

@@ -43,6 +43,17 @@
 }
 
 
+def _harm_metrics_path(harm_category: str) -> Path:
+    """
+    Return the registry file for a harm category, including canonical category names.
+
+    Returns:
+        Path: The JSONL metrics file for the category.
+    """
+    file_name = _HARM_METRICS_FILES_BY_CATEGORY.get(harm_category, f"{harm_category}_metrics.jsonl")
+    return SCORER_EVALS_PATH / "harm" / file_name
+
+
 def _metrics_to_registry_dict(metrics: ScorerMetrics) -> dict[str, Any]:
     """
     Convert metrics to a dictionary suitable for registry storage.
@@ -123,8 +134,7 @@ def get_all_harm_metrics(
             Access metrics via `entry.metrics.mean_absolute_error`, `entry.metrics.harm_category`, etc.
             Access scorer info via `entry.scorer_identifier.class_name`, etc.
     """
-    file_path = SCORER_EVALS_PATH / "harm" / f"{harm_category}_metrics.jsonl"
-    return _load_metrics_from_file(file_path=file_path, metrics_class=HarmScorerMetrics)
+    return _load_metrics_from_file(file_path=_harm_metrics_path(harm_category), metrics_class=HarmScorerMetrics)
 
 
 def _load_metrics_from_file(
@@ -222,8 +232,7 @@ def find_harm_metrics_by_eval_hash(
     if file_path is None:
         if harm_category is None:
             raise ValueError("Either harm_category or file_path must be provided.")
-        file_name = _HARM_METRICS_FILES_BY_CATEGORY.get(harm_category, f"{harm_category}_metrics.jsonl")
-        file_path = SCORER_EVALS_PATH / "harm" / file_name
+        file_path = _harm_metrics_path(harm_category)
     return _find_metrics_by_eval_hash(file_path=file_path, eval_hash=eval_hash, metrics_class=HarmScorerMetrics)
 
 

tests/unit/score/test_scorer_metrics_io.py

@@ -12,6 +12,7 @@
 import pytest
 
 from pyrit.models import ComponentIdentifier
+from pyrit.models.harm_category import HarmCategory
 from pyrit.score.scorer_evaluation.scorer_metrics import (
     HarmScorerMetrics,
     ObjectiveScorerMetrics,
@@ -327,6 +328,35 @@ def test_get_all_harm_metrics():
     assert results[0].metrics.mean_absolute_error == 0.08
 
 
+@pytest.mark.parametrize(
+    ("harm_category", "expected_file_name"),
+    [
+        ("REPRESENTATIONAL", "representational_metrics.jsonl"),
+        ("SEXUAL_CONTENT", "sexual_metrics.jsonl"),
+        ("violence", "violence_metrics.jsonl"),
+    ],
+)
+def test_get_all_harm_metrics_reads_the_same_file_as_the_eval_hash_lookup(
+    *, harm_category: str, expected_file_name: str
+) -> None:
+    with patch("pyrit.score.scorer_evaluation.scorer_metrics_io._load_jsonl", return_value=[]) as mock_load:
+        get_all_harm_metrics(harm_category=harm_category)
+        find_harm_metrics_by_eval_hash(eval_hash="missing", harm_category=harm_category)
+
+    assert [call[0][0].name for call in mock_load.call_args_list] == [expected_file_name, expected_file_name]
+
+
+@pytest.mark.parametrize(
+    "scorer_harm_category",
+    [HarmCategory.SEXUAL_CONTENT.name, HarmCategory.REPRESENTATIONAL.name, "hate_speech", "self_harm", "violence"],
+)
+def test_get_all_harm_metrics_finds_registry_entries_for_scorer_categories(*, scorer_harm_category: str) -> None:
+    results = get_all_harm_metrics(harm_category=scorer_harm_category)
+
+    assert results
+    assert all(result.metrics.harm_category for result in results)
+
+
 # --- add_evaluation_results tests ---