SimInter, posted to arXiv on 13 October 2026, fills out user profiles by running a synthetic interview between two LLM agents, then reports filtered aspect entropy of 1.619 against 0.859 for a statistics-only baseline on a matched set of 100 users. The interviews are grounded in the user’s statistical profile and past conversation, and an audit agent flags fabrication and stereotyping turn by turn. So the simulator talks a user into existence, then talks to that user.