Depth peeling
A multi-pass technique for correct transparency that peels off one visible layer at a time, nearest to farthest.
Transparent surfaces must be blended in depth order, but sorting see-through triangles per pixel is impossible with an ordinary z-buffer. Depth peeling solves it bluntly: each render pass admits only the nearest fragment lying behind the previous pass's depth, so the layers are "peeled off" one at a time like the onion they resemble. Cass Everitt described the technique at NVIDIA in 2001.
The result is exact but expensive — N transparent layers need up to N render passes over the geometry. Later variants like dual depth peeling peel from both ends at once, but on modern hardware order-independent transparency using rasterizer ordered views or per-pixel lists has largely replaced it. It survives as the reference for the "right" answer.