2026-11-03

A century after Gaudi's death, and 144 years into construction, the Sagrada Familia finally got its main spire this February: the tallest church in the world, and one of the strangest. Asterisk takes the obvious next question seriously. If almost everyone finds Art Nouveau beautiful, why did a style that ran from roughly 1890 to 1910 stay a minority taste instead of becoming the default language of the built world? The essay works through the competing explanations - the labour cost of hand ornament, materials, the economics of speculative construction, the movement's refusal to be modularised - and tests them against the fact that only a handful of cities (Riga, Brussels, Vienna, Budapest, Nancy) ever went deep. Along the way it dismantles the 'trad' internet story that modernism simply talked everyone out of beauty, and lands on a less flattering answer: taste didn't change because of an argument, it changed because of a balance sheet.

Deep Blue beat Garry Kasparov in 1996 using 480 custom chess chips and several million dollars of IBM research money. Carlini asks whether any of that was necessary, then answers it empirically: he took a 1995-vintage 200 MHz Pentium Pro with 64MB of RAM, ported Stockfish to it, and measured roughly 3300 Elo at classical time controls - about 500 points above Kasparov's peak, on a machine that fits under a desk. The point is not retro trivia. It is a clean demonstration that in chess, and probably elsewhere, algorithmic progress compounds faster than brute-force hardware, so the historical record of 'the first machine to beat a human' measures sponsorship and engineering effort as much as capability. You can play the Pentium yourself over Lichess, which is a much better joke than any write-up could manage.

Negative results almost never make it into Nature, but this one did: an international collaboration tested three standard ways of measuring cognitive reappraisal - the mental move at the heart of cognitive behavioural therapy, where you reinterpret a situation to blunt its emotional charge - and found they don't agree with each other. That is awkward, because reappraisal capacity is treated as a stable trait tied to resilience, and a large brain-imaging and clinical literature rests on questionnaires assumed to index it. If the instruments aren't measuring the same construct, a lot of that work is measuring something narrower than it claims. It is methodology as news: nothing was discovered except that the field's ruler is crooked, which is often the more consequential finding.

A photographer took the depth map that normally hides inside a lens-blur filter and made it the subject. A diffusion transformer infers the distance of every point in a scene from a single frame; mapped to tone or colour, the result is a sculptural relief of the photograph - 'drawing depth', as the coinage has it. The interesting claim is what happens to focus. Once depth is data rather than a lens setting, the sharp plane is no longer fixed at capture: you can bring several distances forward at once, or isolate one by its value and let everything else fall away. It is a small, well-executed instance of a bigger shift - computational passes long treated as plumbing becoming authorial material - and the side-by-side slider makes the argument better than any essay could.

Two weeks ago Maps Mania proclaimed the age of the pixel map; this week it buries it. The successor is the voxel map: cubes instead of squares, so the scene rotates freely rather than offering four fixed viewpoints. The exemplar is Piotr Migdal's Warsaw in Voxels, which renders OpenStreetMap data with a Minecraft and MagicaVoxel aesthetic in Three.js and turns a city centre into something you can walk around from any angle. Under the joke is a real cartographic argument: the choice of primitive - point, pixel, voxel - silently constrains what a map is able to show, and hobbyists are now iterating on that choice faster than most mapping institutions do.

Thru-hiking blogs are usually logistics, and this one is too - a short mileage day after Whitney, laundry in a cold creek, a game of frisbee, a cough that won't quit. What it carries underneath is the thing the title promises: word reaching a father mid-trail that his son has stage four colon cancer. The mundanity is what makes the dread legible; the day keeps happening around the news. It is also the best short account you'll read of something long-distance hikers discuss and rarely write down well: the trail strips away your information environment exactly when you might need it, and 'remote' stops being romantic the moment something goes wrong at home. Read it as writing about attention and obligation, not gear.

Charles Tandy answers a question almost nobody asks: how did a leathercraft hobby supplier end up selling the first mass-market home computer? Farquhar traces the path - a ten-year-old selling leather scraps to classmates, a wartime pivot when shoe rationing nearly killed the family firm, mail-order craft kits pitched from a Navy hospital bed, and eventually the acquisition of a small Boston hobby chain called Radio Shack. Out of that came the TRS-80 in 1977, which put a computer in ordinary retail stores years before IBM took the category seriously. It is business history with the useful property of being counterintuitive: the PC industry's first retail success was run by a leather salesman who understood hobbyists because he had been one.

China has now published a schedule for putting astronauts on the Moon before 2030, and the detail worth noticing is what it leaves out: no equivalent of Apollo 8 or Artemis II, no crewed flight around the Moon ahead of the landing attempt. That is a different risk posture from NASA's, and it says a lot about how each program sequences crewed versus uncrewed demonstrations. Ars treats the announcement as a concrete commitment rather than an aspiration, which matters because schedules can be compared and slippage can be measured. If both programs hold, the late 2020s become the first period since 1972 with two countries capable of landing people on another world - and the first ever with two independent, non-cooperating ones.

A compound that relieves pain without suppressing breathing, and that appears to reset opioid tolerance without triggering withdrawal, is about as close to a miracle as pharmacology gets - and it has never been tested in humans. Wired follows SR-17018 from a 2017 Cell paper in mice to Reddit and Discord, where people with opioid use disorder have been buying it online and reporting that it let them quit. Then the DEA placed it in Schedule I alongside heroin, cutting off the legal supply and pushing the same users back into grey markets. It is a compact case study in how scheduling decisions, patent incentives, and desperate self-experimentation interact - and why 'not FDA approved' and 'not worth studying' are very different claims.

A twelve-year-old antitrust fight over coffee pods survived a motion to dismiss, and the 139-page opinion is more interesting than the product. After the core K-Cup patent expired in 2012, Keurig allegedly held its dominant position not by brewing better coffee but by contract: agreements with Starbucks, Dunkin', Peet's and Caribou that constrained how those roasters could sell competing pods, plus supplier and distributor terms that limited rivals. The judge found enough evidence for a jury to decide whether those deals foreclosed competition from the very brands customers wanted. It is a clean illustration of where market power goes once a patent lapses - the lock, not the recipe, becomes the moat.