diff --git a/litellm/router_utils/pre_call_checks/prompt_caching_deployment_check.py b/litellm/router_utils/pre_call_checks/prompt_caching_deployment_check.py
index da74788c92b..514fbd18e8c 100644
--- a/litellm/router_utils/pre_call_checks/prompt_caching_deployment_check.py
+++ b/litellm/router_utils/pre_call_checks/prompt_caching_deployment_check.py
@@ -98,6 +98,7 @@ class PromptCachingDeploymentCheck(CustomLogger):
                 request_kwargs.get("system") if request_kwargs is not None else None,
             )
 
+            # A cache miss must leave all healthy deployments eligible.
             model_id_dict: Final = await prompt_cache.async_get_model_id(
                 messages=affinity_messages,
                 tools=(
diff --git a/litellm/router_utils/prompt_caching_cache.py b/litellm/router_utils/prompt_caching_cache.py
index e8ef3d0d3fd..5ef5f050deb 100644
--- a/litellm/router_utils/prompt_caching_cache.py
+++ b/litellm/router_utils/prompt_caching_cache.py
@@ -217,6 +217,8 @@ class PromptCachingCache:
         if cacheable_messages is not None and not cacheable_messages:
             return None
 
+        # Only exact cached-prefix matches can establish deployment affinity.
+        # Partial matches must fall back to normal routing.
         # Use serialize_object for consistent and stable serialization
         data_to_hash: Final = {}
         if cacheable_messages is not None: