We introduce SightSentinel, a system that converts a classroom wall display into a recurring vision screening site. Regular educational material carries brief calibrated optotype probes, and eight specialized agents work together to collect and interpret data. Perception agents recover viewing distance, recognition accuracy, approach behavior, gaze stability, response latency and interocular difference. A quality agent discards records taken under poor geometry, insufficient lighting or inattentiveness. The longitudinal agent accumulates only evidence that deviates from the child’s own baseline. An orchestrator produces a Vision Concern Score and routes it through a safety gate whose output range excludes diagnosis, refraction, prescription or reassurance. The key question is whether many cheap, noisy, well‑gated encounters can achieve a referral decision earlier or more reliably than a scheduled test. The paper formalizes the problem, describes the architecture, outlines a four‑stage validation protocol against clinical reference standards, and specifies conditions under which the approach should be rejected.
Review