Gravitational microlensing happens when a foreground object passes in front of a distant star and its gravity briefly magnifies the starlight. Because the effect depends on mass rather than light, it is one of the few ways to detect objects that emit nothing at all, including free-floating planets and black holes. Automated survey pipelines are built around the standard single-lens brightening pattern. Light curves that depart from that pattern, often the most scientifically interesting ones, are the hardest for software to recover. DISCORD pairs a neural network with human classifiers. The network scores real light curves from the Optical Gravitational Lensing Experiment, and the ones it cannot confidently call are routed to volunteers. Where volunteers agree, that verdict sharpens the model. Where volunteers genuinely disagree, that disagreement is itself treated as a signal that the curve is ambiguous and deserves a closer look. No astronomy background is needed. A short guided tutorial teaches the three shapes that matter, and a four-curve practice set unlocks the review queue. Most curves take under a minute to judge. You answer a few plain questions about what you see, and the system handles the rest. Every classification is real research on real survey data, contributing to a study of how human disagreement can improve automated detection.