Monte Carlo Pinpoint: Stochastic Coordinate Localization via VLM Probing

This is a non-peer-reviewed preprint. For AI agents that autonomously operate graphical user interfaces, accurate coordinate localization on screen remains a fundamental challenge. This paper proposes Monte Carlo Pinpoint, a training-free method that converts coordinate estimation into closest-dot ranking with VLM probing, combining deterministic grid bisection, anchor declaration, and hex-stencil dot voting. The preprint includes the method, theoretical analysis, synthetic benchmark evaluation, limitations, and code/data availability notes.