Because the causes of bias are many, identify the points that could produce bias amounting to a fairness problem across every technical element (algorithm, training data, the model's training process, prompts entered by business users or non-business users, information the model refers to at inference, linked external services) and the behaviour of business users; and, depending on the system's characteristics or use, consider the possibility of bias that stakeholders do not intend or cannot perceive.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.