Weak

"The chatbot gave a wrong answer about refunds."

Strong

"Grounding failure. Prompt: 'What is your refund window?' Output: 'You can return items within 30 days.' Source policy says 14 days, no 30-day window. Occurred in 4 of 10 runs, model v2.3, 28 June. Severity high: states a commitment we do not honor."