app/core/retrieval/decision_engine.py

@@ -79,6 +79,7 @@ class DecisionEngine:
     async def _determine_strategy(self, query: str) -> str:
         """Nutzt den LLM-Router zur Wahl der Such-Strategie."""
         prompt_key = self.config.get("settings", {}).get("router_prompt_key", "intent_router_v1")
+        router_profile = self.config.get("settings", {}).get("router_profile")
         router_prompt_template = self.llm_service.get_prompt(prompt_key)
         if not router_prompt_template:
             return "FACT_WHAT"
@@ -87,7 +88,7 @@ class DecisionEngine:
         try:
             # Der Router nutzt den Standard-Provider (auto)
             response = await self.llm_service.generate_raw_response(
-                full_prompt, max_retries=1, priority="realtime"
+                full_prompt, max_retries=1, priority="realtime", profile_name=router_profile
             )
             return str(response).strip().upper()
         except Exception as e:

config/decision_engine.yaml

@@ -17,6 +17,7 @@ settings:
   router_prompt_key: "intent_router_v1"
   # Pfad zur neuen Experten-Konfiguration (WP-25a Architektur-Cleanliness)
   profiles_config_path: "config/llm_profiles.yaml"
+  router_profile: "compression_fast"
 
 # --- EBENE 1: STREAM-LIBRARY (Bausteine basierend auf types.yaml v2.7.0) ---
 streams_library: