Jev Medicine

A comparative study of system-one decision models across 15 medical datasets and 8,742 instances, measuring accuracy, decision stability, and abstention. Jev rivals leading models on text-driven tasks and is weaker on numerically driven ones, while most models still fail to recognize missing answers.

A decision-native interface that represents each medical decision as a Choice, Noul, or Score over a runtime candidate set and returns a calibrated distribution over it. Against a matched generative baseline it cuts calibration error by an order of magnitude, with counting the one task where the baseline leads.

An evaluation of Jev 1.13 on four medical benchmarks: MetaMedQA, PubMedQA, DiagnosisArena-MCQ, and the NEJM Case Challenges, with GPT-6 Sol as reference. All 8,469 requests returned a valid answer; Jev matched GPT-6 Sol on PubMedQA but trailed on diagnostic cases.

An evaluation of Jev as a low-cost judge of agreement with physician-written radiology reports. A single-question configuration reaches Kendall correlations of 0.573 on RadEvalX and 0.398 on RadEvalExpert, and one support question per statement keeps expert agreement while using 43-45% fewer input tokens.