Fix CI: pre-commit auto-fixes + mypy ignore on dynamic adapter kwargs

- trailing-whitespace + end-of-file-fixer applied to committed
  benchmark run artifacts (CSV/JSON in docs/benchmarks/runs/)
- black + isort reformatted the perf-tuning files (no logic change)
- mypy: narrow type: ignore[arg-type] on
  run_benchmark.py:198 'adapter_cls(**adapter_kwargs)' — adapter_kwargs
  is dict[str, object] sourced from argparse and the runtime types
  match each adapter's signature, but mypy can't statically narrow
  across the polymorphic adapter union

Author: JSv4

opencontractserver/benchmarks/adapters/legalbench_rag.py

@@ -244,9 +244,7 @@ def _ensure_loaded(self) -> None:
             tests = payload.get("tests", [])
             if self.paper_sampling:
                 pre_count = len(tests)
-                tests = _paper_sample_tests(
-                    tests, max_per_subset=self.max_per_subset
-                )
+                tests = _paper_sample_tests(tests, max_per_subset=self.max_per_subset)
                 logger.info(
                     "LegalBench-RAG paper-faithful sampling for subset %s: "
                     "%d -> %d tasks (sorted by random(seed=file_path), cap=%d)",

opencontractserver/benchmarks/management/commands/run_benchmark.py

@@ -195,7 +195,10 @@ def handle(self, *args, **options) -> None:
         if benchmark_name == "legalbench-rag":
             adapter_kwargs["paper_sampling"] = options.get("paper_sampling", True)
             adapter_kwargs["max_per_subset"] = options.get("max_per_subset", 194)
-        adapter = adapter_cls(**adapter_kwargs)
+        # mypy can't statically narrow ``adapter_kwargs: dict[str, object]``
+        # against each adapter subclass's specific parameter types. The dict
+        # values are sourced from argparse, which already validated them.
+        adapter = adapter_cls(**adapter_kwargs)  # type: ignore[arg-type]
 
         self.stdout.write(
             self.style.NOTICE(

opencontractserver/documents/signals.py

@@ -194,9 +194,7 @@ def _gc_orphan_structural_set(sender, instance, **kwargs):
     set_id = getattr(instance, "_structural_set_id_at_delete", None)
     if set_id is None:
         return
-    StructuralAnnotationSet = apps.get_model(
-        "annotations", "StructuralAnnotationSet"
-    )
+    StructuralAnnotationSet = apps.get_model("annotations", "StructuralAnnotationSet")
     Document = apps.get_model("documents", "Document")
     if Document.objects.filter(structural_annotation_set_id=set_id).exists():
         return

opencontractserver/pipeline/embedders/openai_embedder.py

@@ -229,7 +229,11 @@ def _embed_text_impl(self, text: str, **all_kwargs) -> Optional[list[float]]:
         except openai.BadRequestError as e:
             logger.error(f"OpenAI API bad request: {e}")
             return None
-        except (openai.RateLimitError, openai.APITimeoutError, openai.APIConnectionError):
+        except (
+            openai.RateLimitError,
+            openai.APITimeoutError,
+            openai.APIConnectionError,
+        ):
             # Transient: re-raise so callers can retry. See the matching
             # block in ``embed_texts_batch`` for the rationale.
             logger.warning(
@@ -304,9 +308,7 @@ def embed_texts_batch(  # type: ignore[override]
         all_kwargs = {**self.get_component_settings(), **direct_kwargs}
         model = all_kwargs.get("openai_embedding_model", s.openai_embedding_model)
         dimensions = int(
-            all_kwargs.get(
-                "openai_embedding_dimensions", s.openai_embedding_dimensions
-            )
+            all_kwargs.get("openai_embedding_dimensions", s.openai_embedding_dimensions)
         )
 
         try:
@@ -336,15 +338,21 @@ def embed_texts_batch(  # type: ignore[override]
             return out
         except openai.AuthenticationError:
             # Permanent: a wrong API key won't fix itself with retry.
-            logger.error("OpenAI API authentication failed (batch). Check your API key.")
+            logger.error(
+                "OpenAI API authentication failed (batch). Check your API key."
+            )
             return None
         except openai.BadRequestError as e:
             # Permanent: malformed input (oversize, bad dimensions, etc.).
             # Returning None prevents celery from burning retries on
             # something that will fail every time.
             logger.error("OpenAI API bad request (batch): %s", e)
             return None
-        except (openai.RateLimitError, openai.APITimeoutError, openai.APIConnectionError):
+        except (
+            openai.RateLimitError,
+            openai.APITimeoutError,
+            openai.APIConnectionError,
+        ):
             # Transient: re-raise so the celery task's autoretry_for=(Exception,)
             # can take over with proper backoff. The OpenAI SDK already
             # absorbed up to OPENAI_CLIENT_MAX_RETRIES of these

opencontractserver/pipeline/parsers/text_chunkers.py

@@ -250,9 +250,7 @@ def chunk(self, text: str) -> Iterator[TextChunk]:
 # NaN and aborts the entire ingest pipeline. Observed in CUAD documents
 # (e.g. JuniperPharmaceuticalsInc_…) where copy-paste artifacts left
 # runs of ``​`` characters between real paragraphs.
-_INVISIBLE_CHARS_RE = re.compile(
-    r"[   -‏
-  -⁠　­]"
-)
+_INVISIBLE_CHARS_RE = re.compile(r"[   -‏
-  -⁠　­]")
 
 
 @register_chunker

opencontractserver/tasks/embeddings_task.py

@@ -483,8 +483,7 @@ def _batch_embed_text_annotations(
 
     # Carve into sub-batches up front so we can fan them out concurrently.
     chunks: list[list[tuple[Annotation, str]]] = [
-        items[i : i + api_batch_size]
-        for i in range(0, len(items), api_batch_size)
+        items[i : i + api_batch_size] for i in range(0, len(items), api_batch_size)
     ]
 
     # ------------------------------------------------------------------ #
@@ -523,8 +522,7 @@ def _embed_one(chunk):
     # Map future -> chunk index for logging/sub-batch numbering.
     with ThreadPoolExecutor(max_workers=max_workers) as executor:
         future_to_idx = {
-            executor.submit(_embed_one, chunk): idx
-            for idx, chunk in enumerate(chunks)
+            executor.submit(_embed_one, chunk): idx for idx, chunk in enumerate(chunks)
         }
         for future in as_completed(future_to_idx):
             idx = future_to_idx[future]
@@ -616,9 +614,7 @@ def _embed_one(chunk):
                         f"Failed to store embedding for annotation {annot.id}: {e}"
                     )
                     result["failed"] += 1
-                    result["errors"].append(
-                        f"Annotation {annot.id}: store failed: {e}"
-                    )
+                    result["errors"].append(f"Annotation {annot.id}: store failed: {e}")
 
 
 @shared_task(
@@ -744,9 +740,7 @@ def calculate_embeddings_for_annotation_batch(
         # Per-embedder ``api_batch_size`` falls back to the global default
         # for embedders that haven't overridden it (and for legacy paths
         # that pass an embedder instance without the attribute).
-        api_batch_size = getattr(
-            embedder, "api_batch_size", EMBEDDING_API_BATCH_SIZE
-        )
+        api_batch_size = getattr(embedder, "api_batch_size", EMBEDDING_API_BATCH_SIZE)
         if text_only_annots:
             logger.info(
                 f"Batch-embedding {len(text_only_annots)} text-only annotations "

opencontractserver/tests/permissioning/test_permissioning.py

@@ -993,9 +993,7 @@ class SetPermissionsIsNewTests(TestCase):
 
     @classmethod
     def setUpTestData(cls):
-        cls.user = User.objects.create_user(
-            username="is_new_test_user", password="pw"
-        )
+        cls.user = User.objects.create_user(username="is_new_test_user", password="pw")
 
     def test_is_new_skips_remove_perm_calls(self):
         """When called with ``is_new=True`` on a fresh object, the function
@@ -1005,9 +1003,7 @@ def test_is_new_skips_remove_perm_calls(self):
 
         new_corpus = Corpus.objects.create(title="brand new", creator=self.user)
 
-        with patch(
-            "opencontractserver.utils.permissioning.remove_perm"
-        ) as mock_remove:
+        with patch("opencontractserver.utils.permissioning.remove_perm") as mock_remove:
             set_permissions_for_obj_to_user(
                 self.user, new_corpus, [PermissionTypes.ALL], is_new=True
             )
@@ -1090,7 +1086,5 @@ def test_is_new_resolves_int_user_id(self):
             self.user.id, new_corpus, [PermissionTypes.ALL], is_new=True
         )
         self.assertTrue(
-            user_has_permission_for_obj(
-                self.user, new_corpus, PermissionTypes.READ
-            )
+            user_has_permission_for_obj(self.user, new_corpus, PermissionTypes.READ)
         )

opencontractserver/tests/test_batch_embedding.py

@@ -464,9 +464,7 @@ def _mock_response(self, status_code, embeddings=None):
             resp.json.return_value = {"embeddings": embeddings}
         return resp
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_successful_batch(self, mock_post):
         """Successful batch returns list of vectors."""
         embedder = self._make_embedder()
@@ -481,9 +479,7 @@ def test_successful_batch(self, mock_post):
         call_kwargs = mock_post.call_args
         self.assertIn("/embeddings/batch", call_kwargs[0][0])
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_client_error_raises(self, mock_post):
         """4xx response raises EmbeddingClientError.
 
@@ -499,9 +495,7 @@ def test_client_error_raises(self, mock_post):
         with self.assertRaises(EmbeddingClientError):
             embedder.embed_texts_batch(["hello"])
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_server_error_raises(self, mock_post):
         """5xx response raises EmbeddingServerError for Celery retry."""
         embedder = self._make_embedder()
@@ -510,9 +504,7 @@ def test_server_error_raises(self, mock_post):
         with self.assertRaises(EmbeddingServerError):
             embedder.embed_texts_batch(["hello"])
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_non_retriable_exception_returns_none(self, mock_post):
         """Non-retriable exception (e.g., builtin ConnectionError) returns None."""
         embedder = self._make_embedder()
@@ -522,9 +514,7 @@ def test_non_retriable_exception_returns_none(self, mock_post):
 
         self.assertIsNone(result)
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_timeout_raises_for_retry(self, mock_post):
         """requests.Timeout re-raises for Celery retry."""
         embedder = self._make_embedder()
@@ -533,9 +523,7 @@ def test_timeout_raises_for_retry(self, mock_post):
         with self.assertRaises(requests.exceptions.Timeout):
             embedder.embed_texts_batch(["hello"])
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_connection_error_raises_for_retry(self, mock_post):
         """requests.ConnectionError re-raises for Celery retry."""
         embedder = self._make_embedder()
@@ -568,9 +556,7 @@ def test_no_service_url_returns_none(self):
         result = embedder.embed_texts_batch(["hello"])
         self.assertIsNone(result)
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_3d_response_squeezed(self, mock_post):
         """3D response array is squeezed to 2D."""
         embedder = self._make_embedder()
@@ -584,9 +570,7 @@ def test_3d_response_squeezed(self, mock_post):
         self.assertEqual(len(result), 2)
         self.assertEqual(len(result[0]), 384)
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_vector_count_mismatch_returns_none(self, mock_post):
         """Mismatched vector count returns None."""
         embedder = self._make_embedder()
@@ -598,9 +582,7 @@ def test_vector_count_mismatch_returns_none(self, mock_post):
 
         self.assertIsNone(result)
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_malformed_200_missing_embeddings_key(self, mock_post):
         """200 response missing 'embeddings' key returns None."""
         embedder = self._make_embedder()
@@ -613,9 +595,7 @@ def test_malformed_200_missing_embeddings_key(self, mock_post):
 
         self.assertIsNone(result)
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_nan_values_handled_per_item(self, mock_post):
         """NaN values in individual embeddings return None for those items only."""
         embedder = self._make_embedder()
@@ -654,9 +634,7 @@ def _mock_response(self, status_code, embeddings=None, body=None):
             resp.json.return_value = {"embeddings": embeddings}
         return resp
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_embed_text_success_1d(self, mock_post):
         """Successful single-text embedding with 1D response."""
         embedder = self._make_embedder()
@@ -669,9 +647,7 @@ def test_embed_text_success_1d(self, mock_post):
         self.assertEqual(len(result), 384)
         mock_post.assert_called_once()
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_embed_text_success_2d(self, mock_post):
         """Successful single-text embedding with 2D response."""
         embedder = self._make_embedder()
@@ -683,9 +659,7 @@ def test_embed_text_success_2d(self, mock_post):
         self.assertIsNotNone(result)
         self.assertEqual(len(result), 384)
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_embed_text_malformed_200(self, mock_post):
         """200 response missing 'embeddings' key returns None."""
         embedder = self._make_embedder()
@@ -695,9 +669,7 @@ def test_embed_text_malformed_200(self, mock_post):
 
         self.assertIsNone(result)
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_embed_text_nan_returns_none(self, mock_post):
         """NaN in single-text response returns None."""
         embedder = self._make_embedder()
@@ -707,9 +679,7 @@ def test_embed_text_nan_returns_none(self, mock_post):
 
         self.assertIsNone(result)
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_embed_text_client_error(self, mock_post):
         """4xx response returns None."""
         embedder = self._make_embedder()
@@ -719,9 +689,7 @@ def test_embed_text_client_error(self, mock_post):
 
         self.assertIsNone(result)
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_embed_text_server_error(self, mock_post):
         """5xx response returns None."""
         embedder = self._make_embedder()
@@ -731,9 +699,7 @@ def test_embed_text_server_error(self, mock_post):
 
         self.assertIsNone(result)
 
-    @patch(
-        "requests.Session.post"
-    )
+    @patch("requests.Session.post")
     def test_embed_text_exception(self, mock_post):
         """Network exception returns None."""
         embedder = self._make_embedder()

opencontractserver/tests/test_benchmarks.py

@@ -699,10 +699,10 @@ def _random_spans(
         return out
 
     def test_recall_matches_upstream_on_random_inputs(self):
-        from opencontractserver.benchmarks.metrics import char_recall_paper
-
         import random as rng_module
 
+        from opencontractserver.benchmarks.metrics import char_recall_paper
+
         rng = rng_module.Random(42)
         for trial in range(200):
             n_pred = rng.randint(0, 30)
@@ -728,10 +728,10 @@ def test_recall_matches_upstream_on_random_inputs(self):
             )
 
     def test_precision_matches_upstream_on_random_inputs(self):
-        from opencontractserver.benchmarks.metrics import char_precision_paper
-
         import random as rng_module
 
+        from opencontractserver.benchmarks.metrics import char_precision_paper
+
         rng = rng_module.Random(43)
         for trial in range(200):
             n_pred = rng.randint(0, 30)