What changed since v1 When I posted the first version of TraceMind, I got one clear piece of feedback: "this is useful but I need to know if my AI is making things up, not just scoring low." So I built hallucination detection. Then while building it I realized I needed a way to compare prompts systematically. So I built A/B testing too. Here's what's new and how I built it. The original problem