diff --git a/backend-compliance/compliance/services/doc_checks/llm_verify.py b/backend-compliance/compliance/services/doc_checks/llm_verify.py
index c14f71b8..c4ed5997 100644
--- a/backend-compliance/compliance/services/doc_checks/llm_verify.py
+++ b/backend-compliance/compliance/services/doc_checks/llm_verify.py
@@ -15,7 +15,9 @@ import httpx
 logger = logging.getLogger(__name__)
 
 OLLAMA_URL = os.getenv("OLLAMA_URL", "http://host.docker.internal:11434")
-OLLAMA_MODEL = os.getenv("OLLAMA_VERIFY_MODEL", "qwen3.5:35b-a3b")
+# P13: qwen3:30b-a3b liefert zuverlaessige JSON-Antworten im Batch-Modus.
+# qwen3.5:35b-a3b lieferte mit format='json' + langem Prompt leere Strings.
+OLLAMA_MODEL = os.getenv("OLLAMA_VERIFY_MODEL", "qwen3:30b-a3b")
 TIMEOUT = 30.0