Avoid immediate same-provider retries after free-pool 429

contextual_orchestrator/orchestrator.py

@@ -11016,6 +11016,11 @@ def attempt(agent: ModelAgent) -> EndpointAttempt[Any]:
                         # incidental wording in an upstream error body (e.g. a
                         # 400 that happens to mention "invalid arguments").
                         decision = classify_provider_transport_failure(exc.retryable)
+                        if exc.provider_status == 429:
+                            # Quota refusal already selected a cooldown above;
+                            # another immediate call to this agent only spends
+                            # the same quota and must not open its health circuit.
+                            decision = replace(downgrade_to_failover(decision), circuit_failure=False)
                     elif isinstance(exc, ProviderResponseError):
                         if allowed_agent_ids is None:
                             raise

tests/test_rate_limit_aware_admission.py

@@ -23,6 +23,7 @@
 import urllib.error
 from contextlib import contextmanager
 from copy import deepcopy
+from dataclasses import replace
 from email.message import Message
 from typing import Any
 
@@ -506,8 +507,9 @@ def _post_chat_completion(port: int, payload: dict[str, Any], token: str):
 def test_http_route_once_waits_out_storm_and_serves_the_request() -> None:
     """orchestrator/free over /v1/chat/completions (route_once) waits out a 429 storm."""
     orchestrator = TaskOrchestrator(
-        _free_route_agents(), tool_retry_attempts=0, rate_limit_wait_seconds=5.0
+        _free_route_agents(), tool_retry_attempts=1, rate_limit_wait_seconds=5.0
     )
+    orchestrator.policy = replace(orchestrator.policy, realtime_judge=False)
     slept: list[float] = []
     orchestrator._rate_limit_sleep = slept.append
     chat_outcomes = QueuedChatOutcomes(