Because many AI systems exist to process data, and a machine learning system's entire life cycle rests on data availability, the organisation operates the data processes the standard enumerates: acquisition from internal or external sources with an assessment of suitability (bias, breadth and representativeness of expected operational input); exploratory analysis for patterns, relationships, trends and outliers; annotation with metadata such as provenance or labels; preparation into a usable form; filtering of unwanted data, whose effects are examined so that unwanted bias is not introduced; normalisation to a common scale; de-identification where personally identifiable information is present; quality checking for completeness and bias, including a check for data poisoning; sampling of a representative subset; augmentation where data is scarce; and labelling, which associates samples with target variables.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.