MAF paragraphs 3.25 to 3.30: take a risk-based view of which features need transparency measures; for explainability, document how training and model selection were done and why, describe design and expected behaviour in product and technical documentation (asking third-party providers for help), use supplementary explanation tools for black-box models, and consider counterfactual or comparison explanations for individuals, accepting that some algorithms (fraud, AML, security) should not be disclosed; where explanation is impractical, document repeatability through repeatability assessments in live settings, counterfactual fairness testing, identifying and handling exceptions in line with policy, and adjusting for drift in time-sensitive data.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.