They were using the same model. Two researchers. Same prompt. Same settings. One received an answer so precise it felt prophetic. The other got fragments—off-topic, impersonal, wrong. They compared notes. Re-ran tests. Even switched machines. The…