The technical people who build and validate the models own the decisions on feature selection, hyper-parameter tuning and performance metric choices; ownership is assigned and responsibilities communicated so development rests on good reasons and human critical judgement can intervene; decisions are supported by quantitative indicators (train and test performance, consistency across sensitive groups, comparison across hyper-parameter sets) and qualitative ones (decisions to mitigate inaccurate predictions); and the appropriate stakeholders and owners review and approve the model once testing, validation and user acceptance rounds have succeeded before it is productionised.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.