Unoptimized Clinical Impact Judge

agent.json

@@ -25,7 +25,7 @@
       ]
     },
     "lm": {
-      "model": "openroutergoogle/gemini-2.5-pro",
+      "model": "openrouter/google/gemini-2.5-pro",
       "model_type": "chat",
       "cache": true,
       "num_retries": 3,

config.json

@@ -1,6 +1,6 @@
 {
-  "task_model": "openroutergoogle/gemini-2.5-pro",
-  "reflection_model": "openrouteranthropic/claude-4-sonnet",
+  "task_model": "openrouter/google/gemini-2.5-pro",
+  "reflection_model": "openrouter/anthropic/claude-4-sonnet",
   "max_tokens": 8000,
   "temperature": 0.1,
   "test_size": 50,

src/llm_judge/cli/run_gepa.py

@@ -75,7 +75,7 @@ def main():
         print(f"Using separate reflection model: {args.reflection_model}")
 
     model_base = (
-        args.provider
+        f"{args.provider}/"
         if args.provider == "openrouter"
         or args.provider == "bedrock"
         or args.provider == "ollama_chat"