test(e2e): require a 200 inside the regenerate grace window and drop the helper docstrings

tests/e2e/access_control/test_access_control_e2e.py

@@ -76,8 +76,6 @@ class TestAccessControl:
     def test_llm_api_routes_group_grants_every_llm_endpoint(
         self, client: AccessControlClient, resources: ResourceManager
     ) -> None:
-        """allowed_routes=["llm_api_routes"] names a route group, not a path: one
-        entry must open every LLM endpoint while the management routes stay shut."""
         key = client.llm_only_key()
         resources.defer(lambda: client.delete_key(key))
 
@@ -89,7 +87,9 @@ class TestAccessControl:
             f"200 must carry a real completion, not an error envelope: {chat.body[:300]}"
         )
 
-        embedding = unwrap(client.proxy.embed(key, EmbedBody(model=EMBEDDING_MODEL, input=f"route group {unique_marker()}")))
+        embedding = unwrap(
+            client.proxy.embed(key, EmbedBody(model=EMBEDDING_MODEL, input=f"route group {unique_marker()}"))
+        )
         assert embedding.model, f"llm_api_routes key reached /embeddings but got no model back: {embedding}"
 
         denied = client.create_model_status(key, f"e2e-route-group-{unique_marker()}")

tests/e2e/management/test_key_management_e2e.py

@@ -90,8 +90,6 @@ def _is_budget_block(outcome: StreamingResponse) -> bool:
 
 
 def _spend_until_budget_blocks(client: ManagementClient, key: str) -> None:
-    """Drive paid calls until the key's max_budget refuses one. The first call spends,
-    the reservation counter trips the cap, and the next call is the 429."""
     for _ in range(40):
         outcome = client.chat_status(key, SPEND_MODEL, f"spend {unique_marker()}")
         if _is_budget_block(outcome):
@@ -105,8 +103,6 @@ def _spend_until_budget_blocks(client: ManagementClient, key: str) -> None:
 
 
 def _settled_spend(client: ManagementClient, key: str) -> float | None:
-    """The key's recorded spend once it is positive and unchanged across two reads a
-    poll interval apart, so no batched spend write is still in flight when we reset."""
     first = client.proxy.key_info(key).spend or 0.0
     time.sleep(client.proxy.poll_interval)
     second = client.proxy.key_info(key).spend or 0.0

tests/e2e/management/test_management_e2e.py

@@ -12,6 +12,7 @@ from __future__ import annotations
 import math
 import time
 from collections.abc import Callable
+from typing import Final
 
 import pytest
 
@@ -377,12 +378,12 @@ class TestKeyRegeneration:
 
         new_key = client.regenerate_key(old_key, grace_period=REGENERATE_GRACE_PERIOD)
         resources.defer(lambda: client.proxy.delete_key(new_key))
-        revoke_at = time.monotonic() + REGENERATE_GRACE_SECONDS
+        revoke_at: Final = time.monotonic() + REGENERATE_GRACE_SECONDS
         assert new_key != old_key, "regenerate returned the same key string, so no rotation happened"
 
         def old_accepted() -> bool | None:
             outcome = client.chat_status(old_key, "gpt-5.5", f"say hi {unique_marker()}")
-            return True if outcome.status_code != 401 else None
+            return True if outcome.ok else None
 
         _ = _poll(client, old_accepted, "old key was rejected 401 inside its grace period at the deadline")
         assert time.monotonic() < revoke_at, (