People delete their failures, which is why robustness benchmarks still lean on synthetic corruptions applied to clean images. We asked contributors to upload the deletions instead: the blurred, tilted, half-cut-off frames from their own camera rolls, with the one rule that a person can still tell what the subject was. Artificially blurred or corrupted files are refused. Files arrive as the phone saved them, from a global base strongest in Southeast Asia, the United States and Europe. There are no degradation labels; blur type and framing error are easy to label, and the camera-original set, which overlaps heavily in contributors and devices, is an approximate clean reference.