This document discusses machine learning model comparison and evaluation. It describes how the rendezvous architecture in MapR makes evaluation easier by collecting metrics on model performance and allowing direct comparison of models. It also discusses challenges like reject inferencing and the need to balance exploration of new models with exploitation of existing models. The document provides recommendations for change detection and analyzing latency distributions to better evaluate models over time.