MAF paragraphs 3.22 to 3.24: models should have an internal owner; teams responsible for data quality, training and selection keep data accountability practices: understanding data lineage (backward, forward and end-to-end) and keeping provenance records, assessing risks of commingled third-party data; ensuring data quality (accuracy, completeness, veracity, recency, relevance, integrity of joined data, usability, human interventions such as labelling); minimising inherent bias such as selection (omission, stereotype) and measurement bias through heterogeneous and complete datasets; using separate training, test and validation data and testing across demographic groups; and periodically reviewing and updating datasets, alert to reinforcement bias. These apply even to non-personal or anonymised data.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.