It turns out LLMs have strong priors over character names that are model-specific and version-specific. If you find Elena Vasquez and Marcus Chen together on a website, there's a good chance Claude generated it. We stumbled on this as a side finding while working on a model diffing method (CDD), and