In verification and validation the team tests the model's capacity for human involvement and control, considers unintended applications and abuse by malicious actors and takes steps to prevent them, checks the system does what it should with no harm to people, other living beings or the environment, ensures it can indicate how likely unavoidable errors are, pilots at small scale, tests reliability and reproducibility where applicable, runs an explicit evaluation process, and confirms IT security procedures were followed. It follows privacy processes, tests whether outputs can be explained where required and documents how the model was tested for performance and risks for downstream users, keeps testing for bias, engages stakeholders, tests accessibility and universal design, assesses wellbeing and societal impact, and maintains audit trails, impact assessments and the ability to report on and respond to outcomes.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.