So-called "black-box AI" can lead to bad decisions, with little post-mortem capability that would allow stakeholders to determine points of failure. To generate trust in such systems, we must expose the path between data and actionable information or indeed action itself, as is the case in fully automated architectures