We aim to design a high fidelity collection process that primarily aims to design human interaction into the system to do two things: first, diagnose and/or fix problems that the system cannot address automatically, and second, gather extra sensory observations, such as physical samples, that provide contextual information useful for determining data integrity and easing data analysis. We believe that carefully integrating human input into the collection path will facilitate collection of datasets that are more useful than completely autonomously collected data.