rc/SIENTIAPDE-2166- feat: rich metadata (filter/policy/counter) for mlflow_content_gate

tests/laborious/activities/test_gates.py

@@ -6,7 +6,13 @@ from sientia_do.notifications.models import NotificationLevel
 from sientia_do.observability.metrics_controller import MetricsController
 from sientia_do.observability.sientia_monitoring import SientiaMonitoring
 
-from laborious.activities.gates import Gates, input_filter_counters, input_filter_functions
+from laborious.activities.gates import (
+    Gates,
+    input_filter_counters,
+    input_filter_functions,
+    mlflow_content_filter_counters,
+    mlflow_content_filter_functions,
+)
 
 
 @fixture(autouse=True)
@@ -469,7 +475,7 @@ def test_mlflow_content_gate_invalid_filter(gates_activity):
     result = gates_activity.mlflow_content_gate(input_data)
 
     # Assert
-    assert result == (None, 0, '')
+    assert result == (None, 0, '', None)
 
 
 @patch('laborious.activities.gates.mlflow_content_filter_functions')
@@ -491,7 +497,7 @@ def test_mlflow_content_gate_filter_exception(mock_mlflow_content_filter_functio
     result = gates_activity.mlflow_content_gate(input_data)
 
     # Assert
-    assert result == (None, 0, '')
+    assert result == (None, 0, '', None)
     gates_activity.debug.assert_called()
     gates_activity.send_notification.assert_called_once_with(
         metadata=metadata['metadata'],
@@ -517,12 +523,12 @@ def test_mlflow_content_gate_no_filters(gates_activity):
     result = gates_activity.mlflow_content_gate(input_data)
 
     # Assert
-    assert result == (None, 0, '')
+    assert result == (None, 0, '', None)
     gates_activity.debug.assert_called()
 
 
 def test_mlflow_content_gate_with_filter(gates_activity):
-    # Arrange
+    # Arrange: 3 rows, all-NaN after the filter's own normalization — triggers NAN_VALUES.
     input_data = {
         **metadata,
         'filters': {'NAN_VALUES': {'POLICY': 'STOP', 'CONFIG': {}}},
@@ -532,10 +538,15 @@ def test_mlflow_content_gate_with_filter(gates_activity):
     }
 
     # Act
-    result = gates_activity.mlflow_content_gate(input_data)
+    path_flag, confidence, comment, document = gates_activity.mlflow_content_gate(input_data)
 
     # Assert
-    assert result == ('STOP', -1, 'Transformed data not passed the content filter')
+    assert (path_flag, confidence, comment) == (
+        'STOP',
+        -1,
+        'Transformed data not passed the content filter',
+    )
+    assert document == {'filters': [{'name': 'NAN_VALUES', 'policy': 'STOP', 'counter': 3}]}
     gates_activity.debug.assert_called()
     gates_activity.send_notification.assert_called()
 
@@ -554,7 +565,7 @@ def test_mlflow_content_gate_with_filter_not_caught(gates_activity):
     result = gates_activity.mlflow_content_gate(input_data)
 
     # Assert
-    assert result == (None, 0, '')
+    assert result == (None, 0, '', None)
     gates_activity.debug.assert_called()
 
 
@@ -569,10 +580,18 @@ def test_mlflow_content_gate_filter_returns_false(gates_activity):
 
     result = gates_activity.mlflow_content_gate(input_data)
 
-    assert result == (None, 0, '')
+    assert result == (None, 0, '', None)
     gates_activity.debug.assert_called()
 
 
+def test_mlflow_content_filter_counters_covers_every_mlflow_content_filter_function():
+    # Same parity guard as `test_input_filter_counters_covers_every_input_filter_function`
+    # (review finding, 2026-09-30): a filter added to `mlflow_content_filter_functions` without
+    # a matching counter here degrades to counter=0 instead of crashing, but this test still
+    # catches the gap before it ships.
+    assert set(mlflow_content_filter_counters) == set(mlflow_content_filter_functions)
+
+
 def test_get_prediction_store_policy_invalid_policy(gates_activity):
     # Arrange
     prediction_store_policy = 'INVALID_POLICY'

tests/laborious/workflows/subworkflows/test_prediction_process.py

@@ -61,7 +61,7 @@ async def test_run(workflow_mock, prediction_process):
         # mlflow_response_gate (transform)
         ('continue', 0.95, 'Error', None),
         # mlflow_content_gate (transform)
-        ('continue', 0.95, 'Transformed data not passed the content filter'),
+        ('continue', 0.95, 'Transformed data not passed the content filter', None),
         # mlflow_response_gate (predict)
         ('continue', 0.95, 'Error', None),
     ]
@@ -369,7 +369,7 @@ async def test_run_stop_at_mlflow_content_gate(workflow_mock, prediction_process
         # mlflow_response_gate (transform)
         ('continue', 0.95, 'Error', None),
         # mlflow_content_gate (transform)
-        ('continue', 0.95, 'Transformed data not passed the content filter'),
+        ('continue', 0.95, 'Transformed data not passed the content filter', None),
     ]
 
     # Act
@@ -480,7 +480,7 @@ async def test_run_stop_at_mlflow_last_response_gate(workflow_mock, prediction_p
         # mlflow_response_gate (transform)
         ('continue', 0.95, 'Error', None),
         # mlflow_content_gate (transform)
-        ('continue', 0.95, 'Transformed data not passed the content filter'),
+        ('continue', 0.95, 'Transformed data not passed the content filter', None),
         ('continue', 0.95, 'Error', None),  # mlflow_response_gate (predict)
     ]
 
@@ -695,10 +695,10 @@ async def test_path_flag_handler_repeat(workflow_mock, prediction_process):
 async def test_path_flag_handler_repeat_no_document_skips_write_activity_call(
     workflow_mock, prediction_process
 ):
-    # Arrange: no document (e.g. REPEAT triggered by a gate that doesn't build one yet — today,
-    # `mlflow_content_gate`) must not even dispatch the remote write_prediction_metadata
-    # activity — that would be a wasted Temporal round-trip on every REPEAT caused by that
-    # gate, not a rare edge case (review finding, 2026-10-01).
+    # Arrange: no document must not even dispatch the remote write_prediction_metadata
+    # activity — that would be a wasted Temporal round-trip (review finding, 2026-10-01).
+    # Defensive case: all 3 gates now always build a document alongside a non-`None`
+    # `path_flag`, so this guards a caller that doesn't, not today's common case.
     data = {'test': 'data'}
 
     # Act
@@ -1076,7 +1076,7 @@ async def test_run_with_cleanup_prefixes(workflow_mock, prediction_process):
     workflow_mock.execute_local_activity_method.side_effect = [
         ('continue', 0.95, 'ok', None),  # input_gate
         ('continue', 0.95, '', None),  # mlflow_response_gate (transform)
-        ('continue', 0.95, ''),  # mlflow_content_gate (transform)
+        ('continue', 0.95, '', None),  # mlflow_content_gate (transform)
         ('continue', 0.95, '', None),  # mlflow_response_gate (predict)
     ]