Annotation rules must at least cover the annotation objective, data format, annotation method and quality indicators; separate rules must be set for functional annotation and for security annotation, each covering at least the execution and review stages; functional rules must guide annotators to produce data that is truthful, accurate, objective and diverse for the domain concerned; and security rules must steer annotation towards the main security risks found in training data and in generated content, preferably covering all 31 risks in Appendix A.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.