A relevance criterion expresses the relevance property over a domain and needs a technique that can isolate how each individual input contributes, such as symbolic, logical or computational formal methods. The criterion should state the domain measured and the expected results, or at least the methodology for evaluating results where they cannot be set in advance. It may compare architectures or training outputs under the same three conditions as the other criteria: same task, same domain, same objective. Checking may be automated or rest on human judgement, and a human decision can be turned into a new automated test.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.