Verifying physical co-presence is a fundamental requirement in systems such as attendance authentication, access control, and resource auditing. Existing practical approaches often rely on low-dimensional signals such as GPS or Wi-Fi to infer physical presence. However, such signals can be easily shared or manipulated under adversarial conditions, fundamentally limiting the integrity of these systems. We argue that the key challenge of practical co-presence verification lies in preserving the relationship between person, time, and location within the evidence itself. From this perspective, we reformulate attendance authentication as the problem of verifying whether submitted digital evidence faithfully represents a physical attendance state.Based on this insight, we propose SnapAttend, an evidence-based attendance authentication system that jointly encodes person, time, and location into a unified image-based representation. SnapAttend captures a single image containing both the physical environment and a dynamically updated visual token, and binds the resulting evidence to a registered device through hardware-backed signatures. To verify the consistency of identity, time, location, and evidence authenticity under adversarial conditions, SnapAttend employs multiple complementary verification stages, including QR session validation, geometry-aware place verification, recapture detection, and delay-based authenticity analysis. We implement a prototype of SnapAttend and conduct extensive evaluations in real lecture environments with 28 participants. Experimental results show that SnapAttend achieves the lowest FRR (4.76%) among the four evaluated systems while maintaining strong robustness against representative adversarial scenarios, with an overall ASR of 3.85%, representing more than a 17x reduction compared to the strongest baseline system.