An arXiv paper describes a submission to Subtask 1b of ImageEval 2026. The task requires selecting the single visually valid statement from three candidates paired with an image, and it is offered in both English and Modern Standard Arabic (MSA).

The method, per the title, combines constrained selection with rotation-averaged logit scoring for multimodal hallucination detection. The goal is to identify which candidate statement is actually grounded in the image.

The abstract does not include results or comparisons with other systems. It only outlines the task and the general approach.