Detectors trained on clean generator output collapse on images that have been re-encoded by a platform and forwarded through a messaging app. This is that distribution: screenshots of images contributors encountered in their feeds and judged synthetic, each with the platform, the source, a one-line reason for suspicion and a confidence rating. Self-generated images and submissions without source context are refused, and each contributor is capped at five. The labels are human suspicion, not verified provenance; treat them as weak supervision. Unlike WildRF, Chameleon or ITW-SM, it draws on Southeast Asian, American and European feeds together; Kled licenses the screenshots and labels, and the posted images remain third-party content.