Predictability is the property that lets stakeholders make reliable assumptions about the output; users trust a system when they can infer how it will behave even without explaining why, and stop trusting it when it behaves surprisingly in obvious situations. A naive notion based on a human predicting the behaviour is subjective, so predictability should be defined with objective and quantifiable criteria, and trust should be achievable through a statistical guarantee of appropriate behaviour even where precise behaviour cannot be predicted, since many machine learning approaches are inherently unpredictable. Predictability is tied to accuracy, and methods that improve accuracy reduce unpredictable outputs.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.