video1.0<iframe src="https://www.loom.com/embed/bc8d67358b5a407889c873638bc24527" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/sessions/thumbnails/bc8d67358b5a407889c873638bc24527-bd82af44de777eb8.gif545.942In this video, I explain how the judging system on ARB Collective operates, utilizing our AI judge built on Gemini 3 Flash as of February 13th, 2026. Users can submit text or image impressions, with the judge ranking the real target against nine decoy targets based on similarity. I emphasize that the judge is completely blind to the real target, ensuring fair evaluation, and I’ve open-sourced the judging code for transparency. Additionally, I developed an alternative AI judge that, while less effective, still produced results above chance, reinforcing the reliability of our current system. I encourage viewers to explore the open-sourced code and understand the meticulous testing that supports our judging process.