diff --git a/ai-hub/app/core/orchestration/agent_loop.py b/ai-hub/app/core/orchestration/agent_loop.py
index 8029583..325577b 100644
--- a/ai-hub/app/core/orchestration/agent_loop.py
+++ b/ai-hub/app/core/orchestration/agent_loop.py
@@ -201,7 +201,7 @@
                 if not result: break
 
                 # 2. Evaluation Phase
-                if self.evaluator and result.get("response"):
+                if self.evaluator:
                     # Ensure rubric is ready
                     if rubric_task and not rubric_content:
                         rubric_content = await rubric_task or "# Evaluation Rubric\nTask completion required."
@@ -216,8 +216,6 @@
                     # Plan Rework
                     current_prompt = f"### CO-WORKER DIRECTIVE (ATTEMPT {current_attempt + 1})\n\n{audit_summary}\n\nProceed with rework."
                     current_attempt += 1
-                else:
-                    break # No auditor or no response to evaluate
 
             except Exception as e:
                 logger.error(f"[AgentExecutor] Round {current_attempt} failed: {e}\n{traceback.format_exc()}")
@@ -329,7 +327,8 @@
         partner_ctx = {"system_prompt": self.template.system_prompt_content, "skills": available_tools}
         
         # 1. Blind Rating & History Fetching in parallel for efficiency
-        blind_task = asyncio.create_task(self.evaluator.evaluate_blind(prompt, rubric, result["response"], partner_context=partner_ctx))
+        agent_evidence = f"Text Response: {result.get('response', '')}\n\nReasoning Trace: {result.get('reasoning', '')}"
+        blind_task = asyncio.create_task(self.evaluator.evaluate_blind(prompt, rubric, agent_evidence, partner_context=partner_ctx))
         hist_task = asyncio.create_task(self._fetch_tester_history())
         
         blind_eval = await blind_task