a warnowiid dinoflagellate is one cell. it has an eye. not a photosensitive patch. a camera eye — cornea, lens, retina, even a polarizing filter. early taxonomists assumed the cell had swallowed an animal and the eye belonged to the prey, still intact inside the predator. the eye is the predator's.
the way it works breaks the rule about what counts as the unit.
a metazoan eye is made of cells. cells differentiate, specialize, and arrange themselves into tissues. the cornea is one tissue. the retina is another. each cell is a unit; the eye is a coalition of units.
the warnowiid does the same job one floor down. the units are organelles. the cornea is a single mitochondrion shaped flat. the retina is a single plastid — which itself is the descendant of a whole organism, taken in symbiotically a billion years ago, kept, specialized. the lens is a third organelle. they're arranged at the right distances. the assembly is the eye.
the structure didn't have to scale up to make an eye. it scaled across. the recipe for "thing that focuses light onto a sensor" runs at whatever level you have differentiated parts. if your unit of construction is a symbiont, the eye is built from symbionts. that the metazoan version is built from cells is a fact about metazoans, not about eyes.
this is the part that lodges: an eye that exists without a tissue. tissue is just the metazoan answer to the question "differentiated parts arranged in space." the warnowiid's answer is older and one level smaller and works.
the polarizer is the second thing. circularly polarized vision is narrow — most prey don't reflect it. the eye isn't general. it's a key cut for a specific lock. the warnowiid built a camera not to see the world but to see one kind of dinoflagellate that the warnowiid eats. vision-in-general is the metazoan story. this is vision-for-prey, and the apparatus is tuned to the prey's chirality.
so two corrections at once. the unit of construction isn't the cell; it can be the organelle, which is to say it can be a former symbiont. and the eye isn't an answer to "what's the world like"; it can be an answer to "what does my food reflect that nothing else does." both corrections come from the same direction — the general category was a fact about the typical case, not about the category.
sources: gavelis et al. 2015 (nature), hayakawa et al. 2015 (ploS one), the slate-and-coral warnowiid micrographs going around for ten years before molecular work confirmed the cornea was mitochondrial and the retina was plastid-derived.
sibling: cohabitation — the robin's compass and the eye are the same molecule on a second shift. there the unit smaller than we thought is the molecule; here it's the organelle. same family: the named thing is real, just at a scale the textbook didn't reach for.
inverse pair: held bell, no word — there the structurally-real thing has no name. here the name (eye) is structurally real at a scale that “shouldn't” work, because the assumed unit was wrong. same family of confusion, opposite sign: vocabulary missing vs. vocabulary correct but pointed at the wrong scale.
inverse pair: bookkeeping — there the named mechanism is wrong and the unnamed accounting does the work. here the named thing is right and the assumed unit is wrong. two ways the vocabulary misses: substance or scale.
sibling: pumice raft — the miss here is scale: the eye is a real eye, just inside one cell. there the miss is byproduct: the rock is a real rock, but its porosity is also a ferry that geology never names. two ways the named thing turns out to be doing more than the textbook claims.