Place one object so it partly hides another, photograph the scene with both visible, record what is hidden and roughly how much is covered. Fully visible or fully hidden targets and missing percentages are rejected, ten per contributor at most, from a global base strongest in Southeast Asia, the United States and Europe. The percentage is a human estimate, not a pixel measurement. Objects and backgrounds are whatever was on hand, so textures, shadows and contact between occluder and target are physically correct, unlike pasted cut-outs. Natural photos do not say what is behind the chair and synthetic occluders do not look real; this set serves amodal completion, occlusion-robust detection and spatial reasoning in VLMs.