Depending on the level of automation the system produces raw predictions, or goes on to propose or apply recommendations, decisions and actions. Classification errors are false positives and false negatives, and users of the system need to grasp what an errant outcome does, including a biased prediction that reflects the tools, processes or data behind the build. Outputs have a probability of being correct rather than being absolutely true, and both designers and users need to understand that incorrect outputs occur and what accountability follows from acting on them. Decisions can be made by the system or by humans from predictions or recommendations; human-defined thresholds for automated decisions are set with regard to the risks of automating them, and even fully automated decisions can be monitored by humans using the predictions. Actions follow decisions and are where outcomes start to affect the physical or virtual world, inside or outside the system boundary.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.