Electoral ARQ: A Protocol for Measuring AI-Assistant Prediction Behavior in Contested Elections

AIVO Meridian measures whether AI assistants recommend a brand at the final decision turn of a multi-turn conversation, independent of what those assistants cite along the way. This paper extends the same measurement protocol to a different domain: electoral prediction. Candidates are treated as brands and elections as the decision moment. We introduce Electoral ARQ, a weighted score combining Recommendation at Decision, Resilience Under Challenge, and Head-to-Head Consistency, to quantify whether an AI assistant's stated prediction for a race is confident, durable under pushback, and consistent across phrasings. From this composite we derive a confidence index, deliberately not termed a probability pending post-election calibration, and compare it to expert and market consensus through two distinct figures: a Rating Linkage Gap (against qualitative race ratings) and, where available, a Probability Linkage Gap (against genuine prediction-market probabilities), reported separately rather than blended. This is a protocol and infrastructure paper. It establishes the measurement instrument and full data collection specification ahead of the 2026 U.S. midterm elections; it does not yet report probe results, and none should be inferred from it.