fix(interactions): restore provider-scoped model attribution on foreign-pod settlement

Persist the router deployment string in the settlement context and carry it
into the rebuilt logging metadata so reconstruct_model_name yields the same
spend log model string as the create-path billing.

litellm/interactions/background_cost_polling.py

@@ -86,6 +86,7 @@ class BackgroundSettlementContext(BaseModel):
     custom_llm_provider: str
     model: str
     model_group: Optional[str] = None
+    deployment: Optional[str] = None
     litellm_call_id: str
     litellm_trace_id: Optional[str] = None
     call_type: str
@@ -209,11 +210,13 @@ def build_settlement_context(
     metadata = get_litellm_metadata_from_kwargs(kwargs=context.logging_obj.model_call_details)
     reservation_dict = metadata.get("user_api_key_budget_reservation")
     model_group = metadata.get("deployment_model_name")
+    deployment = metadata.get("deployment")
     return BackgroundSettlementContext(
         interaction_id=context.interaction_id,
         custom_llm_provider=context.custom_llm_provider,
         model=str(context.logging_obj.model or context.logging_obj.model_call_details.get("model")),
         model_group=model_group if isinstance(model_group, str) else None,
+        deployment=deployment if isinstance(deployment, str) else None,
         litellm_call_id=str(context.logging_obj.litellm_call_id),
         litellm_trace_id=context.logging_obj.model_call_details.get("litellm_trace_id"),
         call_type=str(context.logging_obj.call_type),

litellm/proxy/spend_tracking/background_settlement.py

@@ -128,7 +128,11 @@ def rebuild_logging_for_settlement(context: BackgroundSettlementContext) -> "Lit
     )
     attribution = {key: value for key, value in context.attribution.model_dump().items() if value is not None}
     reservation = context.budget_reservation.model_dump() if context.budget_reservation is not None else None
-    metadata = attribution if reservation is None else {**attribution, "user_api_key_budget_reservation": reservation}
+    optional_entries = (
+        ("user_api_key_budget_reservation", reservation),
+        ("deployment", context.deployment),
+    )
+    metadata = {**attribution, **{key: value for key, value in optional_entries if value is not None}}
     logging_obj.update_environment_variables(
         litellm_params={
             "litellm_call_id": context.litellm_call_id,

tests/test_litellm/interactions/test_background_cost_polling.py

@@ -390,6 +390,7 @@ def _attributed_litellm_params(reservation: Optional[dict] = None) -> dict:
         "user_api_key_user_id": "user-123",
         "user_api_key_team_id": "team-456",
         "deployment_model_name": "gemini-3-flash-preview",
+        "deployment": "gemini/gemini-3-flash-preview",
     }
     if reservation is not None:
         metadata["user_api_key_budget_reservation"] = reservation
@@ -415,6 +416,7 @@ async def test_poller_persists_settlement_context_with_attribution_and_reservati
     assert context.attribution.user_api_key_team_id == "team-456"
     assert context.model == "gemini-2.5-flash"
     assert context.model_group == "gemini-3-flash-preview"
+    assert context.deployment == "gemini/gemini-3-flash-preview"
     assert context.custom_llm_provider == "gemini"
     assert context.call_type == "acreate_interaction"
     assert context.litellm_call_id == "bg-interactions-call-id"

tests/test_litellm/proxy/spend_tracking/test_background_settlement.py

@@ -39,12 +39,14 @@ INTERACTION_ID = "interactions/bg-foreign"
 def _settlement_context(
     reservation: Optional[SettlementReservation] = None,
     model_group: Optional[str] = None,
+    deployment: Optional[str] = None,
 ) -> BackgroundSettlementContext:
     return BackgroundSettlementContext(
         interaction_id=INTERACTION_ID,
         custom_llm_provider="gemini",
         model="gemini-2.5-flash",
         model_group=model_group,
+        deployment=deployment,
         litellm_call_id="original-create-call-id",
         call_type="acreate_interaction",
         attribution=SettlementKeyAttribution(
@@ -117,7 +119,7 @@ class _FakeRowStore:
 
 @pytest.mark.asyncio
 async def test_rebuilt_logging_bills_with_original_attribution_and_request_id():
-    logging_obj = rebuild_logging_for_settlement(_settlement_context())
+    logging_obj = rebuild_logging_for_settlement(_settlement_context(deployment="gemini/gemini-2.5-flash"))
     response = _response("completed", with_usage=True)
 
     await logging_obj.async_log_background_interaction_completion(result=response)
@@ -134,6 +136,7 @@ async def test_rebuilt_logging_bills_with_original_attribution_and_request_id():
     assert payload["team_id"] == "team-456"
     assert payload["api_key"]
     assert payload["spend"] > 0
+    assert payload["model"] == "gemini/gemini-2.5-flash"
 
 
 def test_rebuilt_logging_start_time_is_naive_for_duration_math():