use-decision-model
A decision model is not a chat model — given a state plus typed questions (noul/choice/score) it returns probabilities and labels, no prose and no rationale, so an agent gets judgments it can compute on (threshold, rank, route) instead of text to parse. USE FOR: gating a yes/no fact, classifying into fixed buckets, scoring quality/severity/risk on a scale, judging replies, filtering prompt injection. Covers crafting state and criteria, sending via the bundled sender, and the verified response signature. DO NOT USE FOR: prose generation, explanations, reasoning traces, or backing a conversational agent.
