ConvApparel introduces a framework to quantify the realism gap between AI user simulators and actual humans. The researchers tested these simulators in a clothing retail context to identify where synthetic users diverge from real behavior. This allows developers to refine simulators for more accurate stress-testing of conversational agents before deployment.