Escalation Tracker

The "NewtonX Mojave" humanities submission guide outlines a framework for creating high-quality, review-ready questions that challenge frontier AI models by requiring substantive reasoning rather than simple recall. Submitters are instructed to ensure questions are narrow, self-contained, and defensible, with rationales and citations that logically support a single correct answer while distinguishing it from plausible alternatives. The guide emphasizes that reviewers should be able to solve questions independently using general domain knowledge and provided information, and it outlines a clear evaluation process where adjudicators resolve disagreements by diagnosing whether issues stem from evaluator error, lack of self-containment, or the need for revision.1