diff --git a/src/tools/__tests__/evaluator.test.ts b/src/tools/__tests__/evaluator.test.ts
index 9372f74..0007799 100644
--- a/src/tools/__tests__/evaluator.test.ts
+++ b/src/tools/__tests__/evaluator.test.ts
@@ -26,22 +26,23 @@ describe('evaluateAnswer', () => {
         const { response } = await evaluateAnswer(
           'What is TypeScript?',
           'TypeScript is a strongly typed programming language that builds on JavaScript.',
+          ['definitive'],
           tokenTracker
         );
-        expect(response).toHaveProperty('is_definitive');
-        expect(response).toHaveProperty('reasoning');
+        expect(response).toHaveProperty('pass');
+        expect(response).toHaveProperty('think');
       });
 
       it('should track token usage', async () => {
         const tokenTracker = new TokenTracker();
         const spy = jest.spyOn(tokenTracker, 'trackUsage');
-        const { tokens } = await evaluateAnswer(
+        await evaluateAnswer(
           'What is TypeScript?',
           'TypeScript is a strongly typed programming language that builds on JavaScript.',
+          ['definitive'],
           tokenTracker
         );
-        expect(spy).toHaveBeenCalledWith('evaluator', tokens);
-        expect(tokens).toBeGreaterThan(0);
+        expect(spy).toHaveBeenCalledWith('evaluator', expect.any(Number));
       });
     });
   });