The argument is an efficiency argument, and that is what makes it interesting rather than merely tidy. If the colors that natural surfaces actually present are restricted compared with the human-made world, you do not need many axes to describe them, and the claim is that four poles arranged as two opposed pairs is the cheapest description that still covers the range [1]. Olshausen's compass analogy is doing real work here: an axis with two mutually exclusive ends carries less than two independent labels would, which is why opponency shows up in a compression story rather than having to be installed by hand [11].
The account's payoff is a division of labor rather than new retinal wiring: Helmholtz's three receptor types describe sensory mechanism in the retina, and Hering's four unique hues describe subjective experience grounded in the structure of the visual environment [7]. Belsten's stronger statement is that no property of light, of the cone photoreceptors, or of neural representation has ever explained why these four hues and not some other set [5]. If that is right, physiology was never going to settle it, and the decisive evidence lives in measurements of the environment and in models; electrodes were never going to be where the answer showed up.
There is a real gap here worth naming directly. The phys.org write-up breaks off mid-sentence exactly as it turns to what Belsten actually did with the problem Olshausen handed him [14]. So the denominator is missing: what was measured, how much of it, sampled from where, and what cost function defined "simplest." Nature's palette is not one fixed quantity; a forest floor in shade and a field of flowers are not the same distribution. The force of the theory rests on that statistic being robust rather than a property of one collection, and that is a question for the paper in the Journal of the Optical Society of America A, not for a summary of it [2].
There is also a small piece of arithmetic worth holding onto. The universal core that Berlin and Kay reported is six terms, and four of them are the unique hues while two, white and black, are not hues at all [9][13]. A theory about the statistics of natural color explains the four; lightness is a separate axis and this account, as described, does not reach it. The six-term target is itself still argued over by linguists [9], so the model is being fitted to a phenomenon that is not fully settled.
My view, with its condition attached: this is the most satisfying answer I have seen to why four rather than three or five, because it is the first one that could be wrong in a specific way. An environment with different spectral statistics should yield different unique hues. The prediction on offer, that other primates with three cone types experience the same opponent hues [10], is the easy case, since the statistics there are the same and so the answer must be. The informative test is the case where the statistics differ. Until someone runs it, what we have is consistency between a model and a 148-year-old puzzle [6], which is a weaker thing than cause.
