chore(lint): weekly black/isort/flake8 sweep

analyzer/management/commands/process_ml_feedback.py

@@ -184,9 +184,7 @@ def show_feedback_statistics(self, options):
             self.stdout.write(f"  {field:<18} {counts}")
 
         # Queries with feedback
-        queries_with_feedback = (
-            histories.values("query").distinct().count()
-        )
+        queries_with_feedback = histories.values("query").distinct().count()
         total_queries = Query.objects.count()
         self.stdout.write("")
         self.stdout.write(
@@ -225,9 +223,7 @@ def _get_queries_to_process(self, options):
         if not options["force_all"]:
             cutoff_date = timezone.now() - timedelta(days=options["days"])
             recent = self._feedback_histories().filter(submitted_at__gte=cutoff_date)
-            queryset = queryset.filter(
-                id__in=recent.values("query")
-            ).distinct()
+            queryset = queryset.filter(id__in=recent.values("query")).distinct()
 
         # Get queries with sufficient feedback
         queries_to_process = []

analyzer/management/commands/train_ml_model.py

@@ -131,9 +131,7 @@ def handle(self, *args, **options):
                     else:
                         # Say why. Silently not deploying is how a bad model
                         # gets mistaken for a deploy that just did not happen.
-                        self.stdout.write(
-                            self.style.WARNING(f"Not deployed: {reason}")
-                        )
+                        self.stdout.write(self.style.WARNING(f"Not deployed: {reason}"))
 
             else:
                 raise CommandError(f"Training failed: {result.error_message}")

analyzer/migrations/0008_mlmodelartifact.py

@@ -1,26 +1,48 @@
 # Generated by Django 4.2.30 on 2026-07-18 05:03
 
-from django.db import migrations, models
 import django.db.models.deletion
 import django.utils.timezone
+from django.db import migrations, models
 
 
 class Migration(migrations.Migration):
 
     dependencies = [
-        ('analyzer', '0007_queryanalysis_schema_insights'),
+        ("analyzer", "0007_queryanalysis_schema_insights"),
     ]
 
     operations = [
         migrations.CreateModel(
-            name='MLModelArtifact',
+            name="MLModelArtifact",
             fields=[
-                ('id', models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')),
-                ('data', models.BinaryField(help_text='joblib-serialized model bytes')),
-                ('byte_size', models.BigIntegerField(default=0)),
-                ('checksum', models.CharField(blank=True, help_text='SHA256 of the stored bytes', max_length=64)),
-                ('created_at', models.DateTimeField(default=django.utils.timezone.now)),
-                ('model', models.OneToOneField(on_delete=django.db.models.deletion.CASCADE, related_name='artifact', to='analyzer.mlmodel')),
+                (
+                    "id",
+                    models.BigAutoField(
+                        auto_created=True,
+                        primary_key=True,
+                        serialize=False,
+                        verbose_name="ID",
+                    ),
+                ),
+                ("data", models.BinaryField(help_text="joblib-serialized model bytes")),
+                ("byte_size", models.BigIntegerField(default=0)),
+                (
+                    "checksum",
+                    models.CharField(
+                        blank=True,
+                        help_text="SHA256 of the stored bytes",
+                        max_length=64,
+                    ),
+                ),
+                ("created_at", models.DateTimeField(default=django.utils.timezone.now)),
+                (
+                    "model",
+                    models.OneToOneField(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="artifact",
+                        to="analyzer.mlmodel",
+                    ),
+                ),
             ],
         ),
     ]

analyzer/ml/core/hybrid_grader.py

@@ -254,9 +254,7 @@ def _load_current_model(self) -> Optional[Any]:
 
             model_data = retrieve_artifact(active_model)
             if model_data is None:
-                model_file_path = os.path.join(
-                    self.model_path, active_model.file_path
-                )
+                model_file_path = os.path.join(self.model_path, active_model.file_path)
                 if not os.path.exists(model_file_path):
                     logger.error(f"Model file not found: {model_file_path}")
                     return None

analyzer/ml/core/training_gates.py

@@ -22,9 +22,7 @@ def real_training_sample_count() -> int:
     """Count TrainingData rows that came from real feedback, not the seed."""
     from ...models import TrainingData
 
-    return TrainingData.objects.exclude(
-        validation_source=SYNTHETIC_SEED_SOURCE
-    ).count()
+    return TrainingData.objects.exclude(validation_source=SYNTHETIC_SEED_SOURCE).count()
 
 
 def real_feedback_gate():

analyzer/ml/core/training_pipeline.py

@@ -264,9 +264,7 @@ def run_training_pipeline(self, force_retrain: bool = False) -> TrainingResult:
                 if ok:
                     self._deploy_model(model_version)
                 else:
-                    logger.warning(
-                        f"Model {model_version} not deployed: {reason}"
-                    )
+                    logger.warning(f"Model {model_version} not deployed: {reason}")
 
             training_time = (timezone.now() - start_time).total_seconds()
 
@@ -640,9 +638,7 @@ def cleanup_old_models(self, keep_versions: int = 5):
                     os.remove(local_file)
                     logger.info(f"Removed old model file: {local_file}")
                 except OSError as e:
-                    logger.warning(
-                        f"Could not remove model file {local_file}: {e}"
-                    )
+                    logger.warning(f"Could not remove model file {local_file}: {e}")
 
             # Remove database record (cascades to MLModelArtifact)
             model.delete()

analyzer/ml/tests/test_alert_notifier.py

@@ -189,13 +189,14 @@ def setUp(self):
         cache.clear()
 
     def test_evaluation_sends_one_email_per_new_alert(self):
+        from django.utils import timezone as dj_timezone
+
         from analyzer.ml.monitoring import alert_evaluator
         from analyzer.ml.monitoring.retraining_system import (
             RetrainingTrigger,
             TriggerReason,
             TriggerUrgency,
         )
-        from django.utils import timezone as dj_timezone
 
         triggers = [
             RetrainingTrigger(

analyzer/ml/tests/test_model_storage.py

@@ -97,9 +97,7 @@ def test_load_current_model_uses_db_artifact_without_local_file(self):
         # came from the database.
         with tempfile.TemporaryDirectory() as empty_dir:
             grader.model_path = empty_dir
-            self.assertFalse(
-                os.path.exists(os.path.join(empty_dir, row.file_path))
-            )
+            self.assertFalse(os.path.exists(os.path.join(empty_dir, row.file_path)))
             loaded = grader._load_current_model()
 
         self.assertIsNotNone(loaded)

analyzer/ml/tests/test_process_ml_feedback_command.py

@@ -145,9 +145,7 @@ def test_selection_threshold_defaults_to_the_collector_threshold(self):
 
         output = self.run_cmd("--stats-only")
 
-        self.assertIn(
-            f"(>= {collector_min} feedback items): 0", output
-        )
+        self.assertIn(f"(>= {collector_min} feedback items): 0", output)
 
     def test_query_with_enough_feedback_is_found_and_processed(self):
         self.add_detailed_feedback(FeedbackCollector().min_feedback_count)

analyzer/ml/tests/test_training_gates.py

@@ -71,9 +71,7 @@ def test_real_count_ignores_seed_even_when_mixed(self):
         self.assertTrue(training_gates.real_feedback_gate()[0])
 
 
-@override_settings(
-    ML_MIN_REAL_FEEDBACK_SAMPLES=3, ML_MIN_TRAINING_SAMPLES=1
-)
+@override_settings(ML_MIN_REAL_FEEDBACK_SAMPLES=3, ML_MIN_TRAINING_SAMPLES=1)
 class TrainingPipelineGateTests(TestCase):
     """The pipeline must refuse synthetic-only data even when the plain
     sample-count gate (ML_MIN_TRAINING_SAMPLES) would pass."""

analyzer/test_analytics.py

@@ -133,8 +133,9 @@ def test_login_fires_user_login_on_the_next_page(self):
         self.assertEqual(response.status_code, 302)
         self.assertIn("_auth_user_id", self.client.session)
 
-        self.assertEqual(self.rendered_event(self.client.get(response["Location"])),
-                         "user_login")
+        self.assertEqual(
+            self.rendered_event(self.client.get(response["Location"])), "user_login"
+        )
 
     def test_logout_survives_the_session_flush(self):
         """logout() flushes the session, then the view writes the flag into
@@ -145,8 +146,9 @@ def test_logout_survives_the_session_flush(self):
         self.assertEqual(response.status_code, 302)
         self.assertNotIn("_auth_user_id", self.client.session)
 
-        self.assertEqual(self.rendered_event(self.client.get(response["Location"])),
-                         "user_logout")
+        self.assertEqual(
+            self.rendered_event(self.client.get(response["Location"])), "user_logout"
+        )
 
     def test_event_fires_exactly_once(self):
         """The pop has to mark the session dirty, or the flag survives and

analyzer/test_seo.py

@@ -131,9 +131,7 @@ def test_home_page_has_exactly_one_self_referencing_canonical(self):
         html = response.content.decode()
 
         self.assertEqual(html.count('rel="canonical"'), 1)
-        self.assertIn(
-            '<link rel="canonical" href="https://querygrade.com/">', html
-        )
+        self.assertIn('<link rel="canonical" href="https://querygrade.com/">', html)
 
     def test_home_page_is_indexable_and_describes_itself(self):
         response = self.client.get("/")

analyzer/urls.py

@@ -9,7 +9,6 @@
 
 # ML Dashboard views (separate module)
 from .ml import dashboard_views
-from .views import ml_alert_views
 
 # Import from modular views package
 from .views import (  # Authentication views; Query grading views; Comparison views; Batch analysis views; History and feedback views; Upload views; Database introspection views; Async processing views; API views; Saved connection views
@@ -39,6 +38,7 @@
     index,
     login_view,
     logout_view,
+    ml_alert_views,
     password_change,
     password_reset_confirm,
     password_reset_request,

analyzer/views/ml_alert_views.py

@@ -20,8 +20,11 @@
 from django.utils import timezone
 from django.views.decorators.http import require_POST
 
-from analyzer.ml.monitoring.rollback import can_rollback
-from analyzer.ml.monitoring.rollback import RollbackError, perform_rollback
+from analyzer.ml.monitoring.rollback import (
+    RollbackError,
+    can_rollback,
+    perform_rollback,
+)
 from analyzer.models import MLAlert, MLModel
 
 logger = logging.getLogger(__name__)

querygrade/settings.py

@@ -140,9 +140,7 @@
 # ship another non-predictive model. This gate counts only rows whose
 # validation_source is not the synthetic seed, so retraining waits for genuine
 # user feedback to accumulate (see #92).
-ML_MIN_REAL_FEEDBACK_SAMPLES = int(
-    os.environ.get("ML_MIN_REAL_FEEDBACK_SAMPLES", "25")
-)
+ML_MIN_REAL_FEEDBACK_SAMPLES = int(os.environ.get("ML_MIN_REAL_FEEDBACK_SAMPLES", "25"))
 
 # Deploy quality gate, read by TrainingConfig. A model must clear BOTH the
 # validation and the held-out test bar, and not show too large a gap between
@@ -152,9 +150,7 @@
 ML_TEST_PERFORMANCE_THRESHOLD = float(
     os.environ.get("ML_TEST_PERFORMANCE_THRESHOLD", "0.7")
 )
-ML_MAX_VALIDATION_TEST_GAP = float(
-    os.environ.get("ML_MAX_VALIDATION_TEST_GAP", "0.15")
-)
+ML_MAX_VALIDATION_TEST_GAP = float(os.environ.get("ML_MAX_VALIDATION_TEST_GAP", "0.15"))
 
 # ML Feature Flags
 # Default OFF: hybrid grading only fires for authenticated users, and the app