style: fix linting and formatting issues

headroom/perf/analyzer.py

@@ -18,7 +18,6 @@ from datetime import datetime, timedelta
 from headroom import paths as _paths
 from headroom.pricing.litellm_pricing import resolve_litellm_model
 
-
 log = logging.getLogger(__name__)
 
 LOG_DIR = _paths.log_dir()

headroom/pricing/litellm_pricing.py

@@ -44,6 +44,8 @@ _MODEL_ALIASES: dict[str, str] = {
 }
 
 _resolved_model_cache: dict[str, str] = {}
+
+
 def resolve_litellm_model(model: str) -> str:
     """Resolve model name to one LiteLLM recognizes, adding provider prefix if needed.
     Results are cached per model name to avoid blocking the event loop
@@ -54,6 +56,8 @@ def resolve_litellm_model(model: str) -> str:
     resolved = _resolve_litellm_model_uncached(model)
     _resolved_model_cache[model] = resolved
     return resolved
+
+
 def _resolve_litellm_model_uncached(model: str) -> str:
     """Uncached resolution — called once per unique model name."""
     if not LITELLM_AVAILABLE:
@@ -82,7 +86,7 @@ def _resolve_litellm_model_uncached(model: str) -> str:
             except Exception:
                 break
     return model
-    
+
 
 @dataclass
 class LiteLLMModelPricing:

headroom/proxy/cost.py

@@ -16,7 +16,6 @@ from typing import TYPE_CHECKING, Any
 
 from headroom.proxy.modes import PROXY_MODE_CACHE
 
-
 if TYPE_CHECKING:
     from headroom.proxy.prometheus_metrics import PrometheusMetrics
 
@@ -567,7 +566,6 @@ class CostTracker:
         self._api_cache_write_1h_by_model.clear()
         self._api_uncached_by_model.clear()
 
-    
     def estimate_cost(
         self,
         model: str,
@@ -595,6 +593,7 @@ class CostTracker:
 
         try:
             from headroom.pricing.litellm_pricing import resolve_litellm_model
+
             resolved_model = resolve_litellm_model(model)
 
             # litellm.cost_per_token handles all token types natively:
@@ -704,6 +703,7 @@ class CostTracker:
             return None
         try:
             from headroom.pricing.litellm_pricing import resolve_litellm_model
+
             resolved = resolve_litellm_model(model)
             info = litellm.model_cost.get(resolved, {})
             cost_per_token = info.get("input_cost_per_token")
@@ -722,6 +722,7 @@ class CostTracker:
             return None
         try:
             from headroom.pricing.litellm_pricing import resolve_litellm_model
+
             resolved = resolve_litellm_model(model)
             info = litellm.model_cost.get(resolved, {})
             uncached = info.get("input_cost_per_token")

tests/test_cost_tracker_counterfactual.py

@@ -37,6 +37,7 @@ def test_savings_at_list_price():
 
     # Savings should be 100k tokens * list input price (NOT affected by cache mix)
     import litellm
+
     from headroom.pricing.litellm_pricing import resolve_litellm_model
 
     resolved = resolve_litellm_model(model)

tests/test_proxy_streaming_resilience.py

@@ -33,7 +33,8 @@ class TestModelResolutionCaching:
         import headroom.pricing.litellm_pricing as lp
 
         with patch(
-            "headroom.pricing.litellm_pricing._resolve_litellm_model_uncached", return_value="anthropic/claude-opus-4-6"
+            "headroom.pricing.litellm_pricing._resolve_litellm_model_uncached",
+            return_value="anthropic/claude-opus-4-6",
         ) as mock_uncached:
             # First call — should invoke uncached resolution
             result1 = lp.resolve_litellm_model("claude-opus-4-6")
@@ -165,7 +166,8 @@ class TestModelResolutionCaching:
         import headroom.pricing.litellm_pricing as lp
 
         with patch(
-            "headroom.pricing.litellm_pricing._resolve_litellm_model_uncached", return_value="resolved/model-a"
+            "headroom.pricing.litellm_pricing._resolve_litellm_model_uncached",
+            return_value="resolved/model-a",
         ) as mock_uncached:
             # Resolve
             result1 = lp.resolve_litellm_model("model-a")
@@ -417,12 +419,12 @@ class TestConcurrentSessionSafety:
             return f"resolved/{model}"
 
         with patch(
-            "headroom.pricing.litellm_pricing._resolve_litellm_model_uncached", side_effect=slow_uncached
+            "headroom.pricing.litellm_pricing._resolve_litellm_model_uncached",
+            side_effect=slow_uncached,
         ):
             # Launch 50 concurrent resolution tasks for the same model
             tasks = [
-                asyncio.to_thread(lp.resolve_litellm_model, "claude-opus-4-6")
-                for _ in range(50)
+                asyncio.to_thread(lp.resolve_litellm_model, "claude-opus-4-6") for _ in range(50)
             ]
             results = await asyncio.gather(*tasks)
 
@@ -503,8 +505,8 @@ class TestConcurrentSessionSafety:
     @pytest.mark.asyncio
     async def test_estimate_cost_concurrent_with_caching(self):
         """Multiple concurrent estimate_cost calls should not block each other."""
-        from headroom.proxy.server import CostTracker
         import headroom.pricing.litellm_pricing as lp
+        from headroom.proxy.server import CostTracker
 
         tracker = CostTracker()