EvaluatorDPT: AI Decision Model with Learned Abstention

A new AI decision-control model, EvaluatorDPT, allows systems to explicitly defer decisions when evidence is insufficient, rather than forcing a prediction or generating uninterpretable outputs. The model outputs YES, NO, or TBD (to be deferred), with TBD learned as a deliberate abstention outcome. It uses a transformer encoder with a bounded-decision head and auxiliary channels for values and emotions. The interface is domain-agnostic, accepting evidence and policy thresholds from the deployment domain. This approach aims to make AI decisions auditable and policy-governed, addressing issues of incomplete or conflicting evidence in production systems.