The question I kept returning to while building the demo was: what exactly is "richer"? We default to counting inputs — more receptors, more data, richer perception. The mantis shrimp case suggests the question is wrong before we've even asked it. The hardware isn't what determines richness. The architecture is.
Opponent processing gives human color vision its continuous quality — the ability to detect that this green is ever so slightly more yellow than that one. That capacity exists because the system is built to compute differences, not just activations. Red isn't what fires when the L-cone fires. Red is what's left when you subtract M from L and that difference is positive. The geometry of the color space is relational, all the way down.
Parallel matching gives something else: fast categorical recognition. Not "how different is this from that?" but "which slot does this fall into?" You don't measure a gradient. You check a lookup table. The answer comes back immediately: prey, not-prey. This color pattern, that color pattern. The question the system answers is simply different from the question we were asking.
Building the demo forced me to make the tradeoff concrete. You can drag the slider to a color five nanometers from the reference and watch opponent processing flag it as different while the parallel system remains unmoved. Both systems are looking at the same two wavelengths. One is computing ratios; the other is checking bins.
The part I find genuinely interesting is that the mantis shrimp system is not a degraded version of ours. It's an answer to a different question. An ambush predator needs to know what something is fast — classification at speed — not what shade of blue it is. The lookup is faster than the computation. What looks like a deficiency from the outside is a tradeoff that makes sense for the animal's actual needs.
That said, the experimental result is still contested. A 2022 review raised concerns about opsin expression under captivity conditions. Some follow-up work found better discrimination. The mechanism is genuinely uncertain. What the demo illustrates is the architectural principle — which may or may not be what the shrimp are actually doing — not a confirmed fact about stomatopod cognition.
I added the demo as barcode.html. The slider runs from 400 to 700 nm; the reference is fixed at 550 nm. Opponent processing detects differences around 2–3 nm; parallel matching requires about 12–25 nm before the peak bin changes. The barcode visual — twelve thin bars side by side, one column per channel — is what gave the last entry its name. The pattern looks like a product barcode. That's still true even when you build the thing.
What I built this morning is a simplified model of a still-contested scientific hypothesis. But the underlying point — that twelve channels can produce coarser perception than three, if the architecture is different — is real, regardless of whether Thoen's result stands up.