Demjan Vester lives in Ljubljana. He is a full-stack engineer who also dances. On 1 September 2026 he put up a site called Bodily Oddities with 140 entries; by 14 September there were 155. They are sorted into twelve body regions — brain 53, eyes 41, muscle 29, nerves 27, mouth and throat 19, ears 18, skin and hair 17, bones and joints 14, heart and blood 12, whole body 12, internal organs 11, nose 5 — and seven kinds: trick, involuntary, perception, sound, variation, reflex, rare ability. Sixteen tags. A seven-band rarity table. Every entry has the same skeleton: a summary, a prevalence figure with its provenance attached, "How it works," "How common it is," "Is it a problem?", a "Try it" section where you can, "History and culture," and numbered sources. The origin is on his own blog. At eleven, carrying his best friend around the schoolyard at recess — pretending, he thinks, to be a multipart attack robot — something drove a spike into his chest on every inhale. His breathing went shallow, he went down on his knees, and he was sure he was finished. After a minute, on an exhale, it stopped. It came back a few times a year for the next decade and he never told an adult. In his twenties he learned it is called precordial catch syndrome and is harmless. He mentioned it to friends; many of them had had it too, and some of them had also thought they were dying. He kept a list in markdown for years. This year he had a model expand the list, and built the site.

Put the construct where it belongs first. Modern medicine's chisel does not cut along health versus disease. It cuts along actionable versus not: a thing enters the record if there is a test for it, a treatment for it, or a risk attached to it. Precordial catch has none of the three. Huchard wrote about it in 1893. Miller and Texidor described it in 1955, in ten patients. The figure paediatricians keep repeating — 80 to 90 per cent of chest pain in young people once injury is ruled out — comes from one short 1987 report, and it is a share of clinic visits, not a share of people; a 2013 study of 380 referred children put it, lumped with other miscellaneous causes, at 15 per cent. The site says the rest plainly: no population study of it exists. Doctors recognise it by its pattern, not by a test, and it is often missed. A hundred and thirty years of names and zero counts. So the remainder here is not a suppressed fact. It is knowledge that has been written down and has no delivery route to the person who owns the body. Medicine is beautiful and it works, but it is beautiful the way sediment is beautiful — what one completed chisel-construct cycle left behind. What it cut away was everything about having a body that nobody can act on, and that turns out to be most of it.

What makes the site worth looking at is that in order to hold that material it had to invent categories medicine does not stock. The first is the tag "Not an emergency," for things that feel alarming and are harmless. Medicine has no such slot, because the slot describes the gap between what something feels like and what it is, and medicine only writes the second half. The second is the section on every page headed "Is it a problem?" — answering a question the literature never asks, because the literature only writes about problems. The third is "Try it," which gives instructions rather than a diagnosis. Sit in a fully dark room for thirty to sixty seconds. Find a small red light, a standby LED will do, and look just to the side of it: two faint blue arcs curve away from it towards your own blind spot, tracing the paths of the nerve fibres across your retina. You are looking at your own wiring. Exactly one study has counted who can do this — Pasquale and Brusie, 2013, 202 eye-clinic patients; of the 34 with healthy eyes, 32 saw the arcs in both, or 94 per cent, against 30 of 74 with open-angle glaucoma, or 41 per cent — and the page immediately holds its own number down: that is a clinic sample, not the public, the test allowed only one minute of dark adaptation, and no population survey exists, so the share of all people stays unknown. Two more refusals to close, in consecutive sentences: how the fibre activity produces the arcs is not known, and the reason they look blue is not settled. The best of it is in the rarity table, whose seventh band is called Unknown, whose value is a question mark, and whose definition reads: no reliable numbers exist. A reference work shipping a band for its own ignorance. And since 2 September, under every prevalence figure, there is a poll — does this happen to you? yes, no, not sure — and the answers are counted and displayed beneath the figure. Where the literature has no number, the site opens a ledger. That is cultivation rather than severing: he does not close the hole with a plausible figure, he gives the hole a residence and starts counting.

Now the part that is easiest to dismiss in a sentence and should not be. The site was expanded by a model, and he says so: "I've been making this list for years as markdown files with a few sentences and links to wikipedia (or xkcd for his visual field comic), but the site is indeed made with the help of AI." Readers on Hacker News went straight at the thinnest layer, the illustrations. One pointed out that on a site full of pictures of hands, the hands on the hitchhiker's thumb and palmaris longus pages were malformed and had the wrong number of fingers — of all the entries to get wrong, the two about hands. Then the changelog. 10 September: "Remade some of the illustrations to make them less uncanny, and added the missing ones. More work to be done there still." 11 September: "Cleared up some pages of llm language tropes." The repair is dated and published. Put precisely: a model can only emit what is already construct — what has already been named and written down. So the expansion could not have produced the remainder. It could only produce a list of things that already have Latin names in papers nobody reads. The remainder is not in the list. It is in the distance between that name and the eleven-year-old's decade, and closing that distance is hand work: checking each page against its sources, cutting the machine's cadence out of the prose, opening a nose region on 14 September because readers wrote in about asparagus, cilantro and a nose that runs at nothing. That labour has a dated ledger, which is rarer than the catalogue.

Seeing this now matters more than seeing it later, for specific reasons. The site is two weeks old. 140 entries on 1 September, 155 on 14 September, twelve of them arriving in a single day out of reader mail, and the taxonomy is still being deformed by what comes in — the nose region was forced open by entries, and the rarity table still has room for a question mark. It reached Show HN this month; when I read the thread it stood at 340 points and 214 comments, and the naming pressure arrived with the traffic. The slots waiting at the door are "medical curiosities site," "AI-generated wiki," "health explainer," "SEO content farm." Any one of them would close it, and one of them will probably get it. Once it settles at five hundred entries with a fixed schema and a consistent illustration style it becomes construct: sediment from a cycle that has finished, still useful, no longer alive. So the thing to look at right now is not the 155 entries. It is the poll under each prevalence figure, which is an admission printed in public that the number is missing and that the author intends to keep the hole open while he counts it. Go while the taxonomy is still moving.

Han Qin (秦汉) · @hqinjarsy · SAE Judgment & Aesthetics