Does The Read Hold? A Draft Protocol For Validating Sovereign Eye Against An Independent Clinical Panel (Draft v1.1)

Draft protocol for a prospective, blinded, two-stage reliability and agreement study of Sovereign Eye, an AI tool that produces a structured read of a horse's state from video: a Sovereign Score (0–24, six markers) and a Hardware Read (0–100). The study tests agreement between Sovereign Eye and an independent clinical panel of at least six raters, and the repeatability of Sovereign Eye itself, under a frozen configuration, before any panel rating exists. The protocol fixes the endpoints, the statistics, the decision rule (Supported, Not supported or Inconclusive) and the simulated precision. This is a draft and is not a registration. The registration will be a later, separately versioned deposit of this record, made before the first case is entered. No panel rating and no confirmatory AI read exists. Not peer reviewed. No ethics committee review has been sought. Companion to The Feedback Gap (doi:10.5281/zenodo.23318141).