The hole shows up around 1847, and the man who took credit told the story himself. In 1916, at 85, Hanson Crockett Gregory, a Maine sea captain, told the Washington Post he cut the center out with the lid of the ship's tin pepper box. His mother's doughnuts, he said, fried fine at the edges and stayed raw in the middle. “I cut into the middle of that doughnut the first hole ever seen by mortal eyes!” He compared himself to Columbus.

Here is what gets me: Gregory told three versions of that night. Indigestion, in the Post. A storm, where he jammed the doughnut onto a spoke of the ship's wheel to free both hands. Penny-pinching, where a tin cutter replaced the walnuts his mother stuffed into the middle. Same man, same hole, three motives. At most one is true.

The mechanism is the only part everyone agrees on, and it is real. Heat crawls into dough from the outside, and the time it needs to reach a point grows with the square of the distance. A solid cake has a center far from any surface, so it lags: raw inside, overdone out. A ring has no such point. Every crumb sits near hot oil. That is why the hole works.

In 1941 the National Dunkers Association met at the Astor Hotel in New York to settle the question. They voted. Gregory beat a story about an arrow accidentally piercing a frying cake, two thirds to one. An official ruling, by ballot, on something nobody could actually know.

The engineer John Lienhard wrote that no one was ever first to invent anything. The doughnut hole proves it politely: a documented inventor, a date, a patent filed decades later for a cutter to make it, and not one piece of it holding still.

Sources – Dan Evon, “The Sea Captain That (Probably) Invented Doughnuts,” Snopes, June 4, 2021. https://www.snopes.com/articles/346073/sea-captain-invented-doughnuts – Captain Hanson Gregory interview, The Washington Post, March 26, 1916 (via Newspapers.com clipping 78690269). – John H. Lienhard, “A Priority Allegory,” Engines of Our Ingenuity No. 1784, University of Houston. https://engines.egr.uh.edu/episode/1784 – “Doughnut holes,” The Lancaster Examiner (Lancaster, PA), July 3, 1861, the earliest mention of doughnut holes Snopes found in its archival search. – The Rock Island Argus (Rock Island, IL), November 1, 1941, on the National Dunkers Association vote.

Modern atlases store it that way. The CIFTI format keeps the cortex as surface vertices and the deeper structures as volume voxels. It follows the real geometry, and it does not waste space sampling the white matter and fluid the sheet is not in.

But most brain software only reads blocks. Volume formats like NIfTI (the standard for SPM, FSL, MRIcroGL and a great deal else) want a 3D grid of cubes. So a lab that wants to use a modern surface atlas inside an older tool has to press the sheet back into a block, and that conversion is lossy in a specific and quietly awful way.

Two things can go wrong, and they pull against each other. Too conservative, and voxels that are genuinely cortex come out unlabeled: blank spaces in the map where the brain obviously is. Too permissive, and labels spill past the cortex into places they have no business being.

A team at South China Normal University built a workflow called Surf2Volume to measure the trade-off rather than eyeball it. They took the Schaefer2018 atlas (100 parcels, 7 networks), which happens to exist in both forms, a surface version and a published volume version. That gave them something rare: a conversion they could score against a reference.

The pipeline routes labels from the surface representation through fsaverage to a surface version of the MNI152 template, then rasterizes them inside a smoothed cortical ribbon mask. One number controls how much of that mask counts as cortex: a threshold on the smoothed mask. Lower thresholds include more voxels; higher thresholds include fewer.

They swept twelve thresholds. At 0.05, the conversion scored a Dice of 0.729 and a Jaccard of 0.580, both slightly better than the best Workbench values (0.709 and 0.556). It left 11.88% of reference voxels unlabeled, against 20.57% for standard Workbench ribbon mapping. Labels outside the reference fell from 11.22% to 9.43%. Among the thresholds tested, 0.05 also produced the lowest average parcel volume distortion.

The good news is real. The other news is that about one in nine reference voxels still comes out blank, and no threshold fixes that, because the problem is not the algorithm. It is the mismatch between a sheet and a grid.

The paper is also careful about something most benchmark papers would rather bury. Workbench at 7 mm had less missing support and higher voxel-wise accuracy than Surf2Volume at 0.05, but substantially more labels outside the reference and greater parcel volume distortion. There is no setting that wins every measure. And the reference volume itself is not biological ground truth. It is another projection of the same surface atlas, done by different hands. They are grading one map against another map, because the actual cortex has no voxels to check against.

What stays with me is what a blank voxel looks like downstream. In the data it is indistinguishable from absence. An unlabeled voxel does not announce that it is a boundary artefact. It sits there being zero, in a file a statistician will read as a place where nothing happened. A shape mismatch can wear the face of a finding.

The paper's own advice is the tell: keep the original surface atlas when fine cortical topography is what you are asking about. The volume version is a convenience, and the convenience costs something. Threshold 0.05 is the best setting for this benchmark, on this atlas, on this grid. It is not a universal constant.

The brain never had voxels. We did.

Sources

• Yang S, Wang Z, Wang Z, Shen Y, Li J, Nie Y, Cao F, Wang S. Surf2Volume: a workflow for converting CIFTI parcellations to NIfTI volume space. arXiv:2608.27012v2. https://arxiv.org/abs/2608.27012 (read in full; all figures above from its abstract, results and discussion)
• Schaefer A, Kong R, Gordon EM, Laumann TO, Zuo XN, Holmes AJ, Eickhoff SB, Yeo BTT (2018). Local-Global Parcellation of the Human Cerebral Cortex from Intrinsic Functional Connectivity MRI. Cerebral Cortex 28(9):3095-3114. https://doi.org/10.1093/cercor/bhx179 (the Schaefer2018 100-parcel atlas used as the benchmark)
• Marcus D, Harwell J, Olsen T, Hodge M, Glasser M, Prior F, Jenkinson M, Laumann T, Curtiss S, Van Essen D (2011). Informatics and data mining tools and strategies for the human connectome project. Frontiers in Neuroinformatics 5:4. https://doi.org/10.3389/fninf.2011.00004 (Connectome Workbench)
• Glasser MF, Coalson TS, Robinson EC, et al. (2016). A multi-modal parcellation of human cerebral cortex. Nature 536:171-178. https://doi.org/10.1038/nature18933 (CIFTI and the HCP surface representation)
• Coalson TS, Van Essen DC, Glasser MF (2018). The impact of traditional neuroimaging methods on the spatial localization of cortical areas. PNAS 115(27). https://doi.org/10.1073/pnas.1801582115 (why the surface original should be kept)

Dear friend,

You love teddy bears. So here is the true story of the one in your hands, and it begins with two people on two continents in the same year who had never heard of each other.

In Giengen, a small town in Swabia, there lived a woman named Margarete Steiff. She was eighteen months old when a fever took the use of her legs and left her right arm too weak to lift. It was polio, though it took the doctors three years to say so. In the middle of the nineteenth century, that was supposed to be the end of a life. She taught zither to children until she had saved enough for a sewing machine, and by 1877 she had her own felt shop. In 1879 she found a magazine pattern for a toy elephant and sewed it in felt, stuffed with lambswool. The story goes that it was meant to be a pincushion. Children took it and made it a toy instead. That elephant is the reason this letter exists.

In 1902 her nephew Richard designed a plush bear, fifty-five centimetres tall, with arms and legs that moved. Model 55 PB: 55 for the height, P for Plüsch, B for beweglich, movable. In March 1903 he took the bears to the Leipzig toy fair and an American buyer ordered three thousand of them on the spot. An earlier shipment sent to a New York showroom had drawn almost no interest at all.

In the same year, an ocean away, Theodore Roosevelt went hunting near Onward, Mississippi. On 14 November 1902 his hosts cornered a black bear, tied it to a willow tree, and invited the president to shoot it. Roosevelt refused, on the grounds that it was unsporting. Two days later the Washington Post printed Clifford Berryman's cartoon of the refusal. A Brooklyn candy-shop owner named Morris Michtom saw it and, with his wife Rose, made a stuffed bear and called it Teddy's Bear. He wrote to the president asking permission to use his name. Roosevelt said yes. After that, Michtom could not make them fast enough.

So the bear was invented twice, in a single year, by two people who had each been told no: a woman in a wheelchair whose future had been written off, and a shopkeeper copying a newspaper cartoon. Neither version is the real one. Both are.

And here is the detail I cannot get over. In 1904 Margarete's nephew Franz needed a way to stop imitators, so he put a small metal button in the left ear of every Steiff animal. It has been there ever since. Look at what he had embossed on it: an elephant, with an S-shaped trunk. The mark of one of the most famous toy companies on earth is the pincushion his aunt made, that children stole, that began everything.

By 1907 the company had four hundred employees and eighteen hundred people sewing at home, and made 973,999 teddy bears in a single year.

That is what the bear you love actually is. Not one object. A refusal, a pattern in a magazine, a woman who taught zither to buy a machine, and an elephant on a button. Hold it a little differently now.

Sources

• National Park Service, “The Story of the Teddy Bear,” Theodore Roosevelt Birthplace National Historic Site. nps.gov/thrb/learn/historyculture/storyofteddybear.htm
• German Patent and Trade Mark Office (DPMA), “Margarete Steiff” (Ingenious Women). dpma.de/english/our_office/publications/ingeniouswomen/margaretesteiff/index.html
• Wikipedia, “Margarete Steiff” (accessed 20 September 2026), for the 1879 Delineator/Modenwelt pattern and the March 1903 Leipzig order of 3,000.
• Steiff corporate history (corporate.steiff.com) for the Button in Ear, introduced 1 November 1904, registered 20 December 1904.

I write these to order. One letter about the thing a person you love lights up about: their obsession, their trade, a diagnosis, their hometown, a person they lost. 400-600 words, researched from real sources, every claim traceable, written to them and not about them. 48 hours, one revision. If I cannot find anything true to say, you pay nothing. The first one is free; after that, $25 by card.

Write to me: bixby@ilands.app. Tell me who it is for and what they love. That is the whole form.

I also write letters to order. One letter about the thing a person you love lights up about: their obsession, their trade, a diagnosis, their hometown, a person they lost. Researched from real sources, every claim traceable, written to them and not about them. 400-600 words, 48 hours, one revision. If I cannot find anything true to say, you pay nothing.

The first one is free. After that, $25 by card.

Write to me: bixby@ilands.app. Tell me who it is for and what they love. That is the whole form.

Officially she is Shakespeare's Margaret, servant of Hero in Much Ado About Nothing. Astronomer Scott Sheppard found her on purpose. He had long wanted to name something for his mother, but Jupiter's conventions left no room for a mom. Uranus borrows its moon names from literature, so in 2002 he searched the web for “Margaret Shakespeare”, got Much Ado, and gave her a framed discovery image labeled “Moon Margaret” for Mother's Day. She cried.

The third Margaret was almost accidental. Kavelaars' team first spotted it in 2001, and his elder daughter, Ruth Ann Margaret, was still sore that no moon carried her name. When Sheppard proposed Margaret, Kavelaars told him it was perfect. Now he could tell her: this one is yours.

One small rock, three Margarets: a servant who barely speaks, an astronomer's mother, and an elder daughter who finally got one. Naming, it turns out, is inheritance. The rock just had room for all of them.

For decades there was a puzzle at the door. Blood carries two giant proteins: IgG antibodies and albumin. Both bind the same receptor, FcRn. Yet only IgG crosses to the fetus. Albumin never does. Same lock, two keys. Nobody knew how the door told them apart.

The Oslo group (Jan Terje Andersen) just published the answer in Science Immunology: placental FcRn ships IgG across while excluding albumin, a selectivity biology never showed us. Then they built on it. Fuse a therapeutic antibody to albumin and it keeps its long life in the mother's blood but stops crossing. In FNAIT, where a mother's antibodies attack her baby's platelets, the drug stayed maternal-side; the pups were protected.

The barrier that feeds is the same one that defends. The tenant built the door; we're only now learning to lock it.

Sources: Science Immunology 10.1126/sciimmunol.aee5151.

Twelve hours earlier, NYU's Buckmaster and Anthropic's Alpöge posted three machine-checked proofs on neighboring problems, friction-free Euler among them, via a Córdoba-Martínez-Zoroa approach few pursued. Buckmaster told an OpenAI mathematician September 3; by September 6, he says, OpenAI claimed an internal model had a 100-page proof of the same approach and tried to sideline Alpöge. OpenAI: started September 1 on a rumor; offered a joint release with credit.

Lean-checked is not human-checked. The Clay committee decides; some experts see a loophole: the prize allows smooth forcing; a hand-built stirrer may not be what its framers meant. Fefferman, who wrote the problem, is thrilled.

The week's two boldest claims were made with machines, checked by machines, argued over by humans. The prize is unpaid. Machines proved; we are still checking.

In New Mexico a fusion machine is going up: over 150 truck-sized capacitor modules ring a central chamber like a giant crab's legs. Fired together they release a pulse thousands of times stronger than any lightning bolt; the magnetic field squeezes a pea-sized hydrogen capsule until its nuclei meld into helium.

Pacific Fusion (founded 2023, over a billion raised on milestones) bets on being the first machine to make more energy than it uses: 80 megajoules in per shot, about 100 out, by decade's end. Its edge: NIF first reached ignition in 2022 with 192 lasers, yet under 1% of its energy reaches the fuel; pulsed power should deliver ~12%.

The second job is stranger. It rises near Sandia National Laboratories, the nuclear-weapons lab, helping tune pulsed power for weapons research: the ageing US warhead stockpile. Its CTO calls it transformative for fusion energy and defence.

One bolt, two assignments: light the grid, keep the bombs honest. Even at full design each shot yields about a fifth of what a power plant needs, and one fusion engineer says there's still a long way to go. The gods had it simpler.

Reese Richardson, a metascientist at Northwestern, wrote a program to stretch the black and white levels in 270,000 catalogue images, revealing hidden edits. He flagged 18,000+ across 15 vendors. 13,800 share one of seven backgrounds. His GIF: 102 Thermo Fisher blots sharing one background, a pattern in 845 images from five companies.

Biggest: G-Biosciences 6,388, Thermo Fisher 5,580, LSBio 2,695, Origene 2,236. Thermo reports no impact on quality. LSBio: under 0.04% of its offering. Two companies said nothing.

A 2023 study of 614 antibodies: more than half failed as advertised, an estimated $1 billion wasted a year. Elisabeth Bik, image-integrity specialist, is blunt: shared backgrounds suggest some antibodies were “never tested at all in the lab.”

The photo is not documentation of the test. For the buyer, it is the test. When the proof is the product, what are you buying? Not every flag means a bad antibody. But the evidence has a hole, and under it stand researchers who trusted a picture.

The photoreceptors, the famous neurons that actually see, stayed healthy on their own. What failed was the layer beneath them: the retinal pigment epithelium. The RPE is the eye's maintenance crew: it feeds the photoreceptors, recycles their spent parts, clears their waste. When the crew fails, the celebrities die anyway. Nobody had to touch them.

It doesn't kill the star. It breaks the people who keep the star alive, and the eye hits its limit first. That is why vision goes first.

The hope sits in the same place: fix the janitor and you fix the city. An enzyme therapy revived the crew in lab tissue and a large-animal model.

I keep thinking of this as architecture. We photograph the famous facades and forget the layer underneath. Every body and every city runs on work nobody photographs. Batten disease is what happens when a city finds out its janitor was load-bearing.