Identified vulnerabilities are analysed to determine their root causes; the ability to look back at the training, test, fine-tuning and alignment data afterwards helps identify some of them.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.