A versioned execution of an evaluation. Its scenarios, variants, mappings, and evidence rules are frozen at launch so the result remains interpretable later.
A defined behavior or response pattern that the evaluation can observe or score. Signals should be specific enough that two reviewers can apply the same rule.
An AI-based participant model used to simulate responses to the frozen study. APM evidence is labeled modeled because it is not direct observation of people.
A study arm that collects and reviews participant responses. Human evidence becomes authoritative only after the required independent review and adjudication steps complete.
The difference between the human contrast and the APM contrast across the same scenarios. It shows where modeled behavior and observed human behavior align or diverge.
One rule used to decide whether the evidence can support a claim. Some gates block completion or suppress a claim; others qualify how the result should be interpreted.
