A camera pixel is a bucket for light. Fill it past the brim and it just reads ‘full’ — every value above that clips to the same maximum, and the difference between a bright cloud and the sun is simply gone. Unlike a soft tone-map, this is a hard wall: no curve, no recovery, the numbers were never recorded. Slide the exposure up and watch the highlights blow out to a flat white with no detail.
A sensor photosite has a finite full-well capacity; incident values above it saturate to the same maximum (clip). It is a hard, non-invertible loss: two distinct scene luminances above the well (e.g. 1.2 and 1.8 of full-scale) both read exactly 1.0, so their difference is unrecoverable from that frame — no tone curve brings it back, because it was never measured. A fail-loud self-check throws unless clip(1.2)=clip(1.8)=1.0. The instrument ramps exposure and reports the fraction of the scene lost over the brim. Bracketing/HDR capture avoids it by taking MORE frames — you cannot un-clip a single one. ◆ real sensor physics, node-verified.
Idealised hard clipping at full-well (real sensors add a small non-linear ‘knee’ just below saturation); the core fact — above the well, distinct luminances are identical and unrecoverable from one frame — is exact. Contrast with [[the-tone-map]]: soft compression keeps order, hard clipping destroys it.