Verification and validation exercise the AI throughout the ODD, at its boundaries and on realistic edge cases, and demonstrate safe behaviour on inputs outside the ODD. When the AI is updated, for example with new training data, testing covers existing and new behaviours through new test cases and regression testing to confirm the update's intent and that no unwanted side effects such as catastrophic forgetting or model drift have crept in. Evidence supports the sufficiency of verification and validation, such as metrics on ODD coverage, coverage of explicit and inferred requirements, structural coverage of algorithm and model, and performance metrics in the tested context.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.