A stability criterion decides whether the stability property holds across the whole stated domain, not merely on the training, validation or test sets, and is checked with the formal methods of 6.2. The criterion must at least state the domain value space, the output value space over which it was measured and the stability property expected. It may serve to compare models, but only when the networks perform the same task, the criterion is applied to the same domain and it proves the same objective. Because it relies on prior knowledge of the expected output, the standard recommends it for any decision-making task handled by a network, such as classification or identification.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.