Light falls on an L, an M and an S cone, and each returns nothing but a count. Adding and subtracting those counts gives three opponent channels. The three channels together, and no one of them alone, specify a color.
What is idealized here. The cones are the Stockman–Sharpe 2° fundamentals (the CIE 2006 physiological observer), converted to quantal units, so they include filtering by the lens and macular pigment; their quantal peaks are at 566, 541 and 441 nm. These are measured curves, not the nomograms of the first simulation, because the swatch needs a colorimetric basis. The absorption probabilities therefore differ slightly between the two: here an L cone catches 0.51 of 620 nm photons, there 0.44. As in the first simulation, absorption probability is set to normalized sensitivity , so a photon at a cone's peak is always caught.
Counts are taken over a one-second window of simulation time, a slowed-down stand-in for the tens of milliseconds a cone integrates over. Photon rates are per cone. Only a sample of photons is drawn; the counts include every one. The reference white is CIE D65 at 1,000 photons per second per cone. Dividing each count by its count for white is the same comparison as the cone contrast of the first simulation: a cone contrast is this relative signal minus one, and the differences between cone classes are identical.
The channel weights are schematic but not arbitrary. L + M is weighted 0.70 and 0.30, the CIE luminance weights for this white. S − (L + M) subtracts the plain average of l and m, as in the first simulation, rather than the luminance-weighted L + M; that keeps it close to, but not exactly on, the cardinal S axis. The channels are graded numbers here; how a ganglion cell carries a signed value as a departure from a maintained rate is shown in the first simulation. Because the three channels are a linear, invertible recombination of the three counts, they carry exactly the same information, no more and no less, which is why the swatch can be computed from them.
The swatch is colorimetry, within limits. The channels are converted back to cone excitations, then to CIE 2015 XYZ through the CIE 170-2 matrix, then to sRGB balanced to D65. It is correct only on a calibrated sRGB screen, for an eye adapted to the white. Every single-wavelength light lies outside what any three-primary screen can show; for those, the primary that would have to be negative is set to zero, a standard approximation, and the line under the swatch says so. Clipping lowers saturation and can also shift hue: from about 495 to 520 nm the swatch is drawn greener than the light, so in the sweep those wavelengths look alike in the swatch while the marker on the map keeps moving. Brightness follows L + M through a compressive function, relative to the brightest version of that color the screen can make, so brightness is faithful within a hue but not exactly across hues. Color appearance also depends on surround, size and adaptation, which this simulation does not model.
Hue words describe swatches, never channels. The four ends of the two chromatic axes look rose-red, blue-green, violet and yellow-green. Only one is near a unique hue: unique red lies close to the rose-red end, while unique green, blue and yellow fall between the axes, and unique red and green are not opposite each other (Webster, Miyahara, Malkoc and Raker 2000). The transformation from retinal opponent signals to experienced hue continues in the thalamus and cortex.