How a topic reaches this page, and what the stamps mean. The path from conversation to card, and the rules that say what to trust.
A topic appears only when at least three independent voices discussed it in the last seven days. Two voices from one house count as one. A mere relay counts as none. Noise (views, shares) lets nothing in; it only decides the order. Today, 30 topics cleared the bar.
More than 100 voices, picked one by one: labs, researchers, people who build, people who measure. And a handful of journals, blogs and repositories. The list itself stays in the drawer.
Every night, everything that was said goes into one large graph: the posts, the articles, the names that keep coming back (models, labs, tools) and the links between them. Algorithms look for communities in it, conversations that answer each other. Each one is then sorted by theme, weighed, and ranked. Those that clear the three-voice bar become cards.
Language models name each topic and summarise it, in three languages. Then they step back: they connect the day’s topics to draw a few ideas out of them, the day in brief. Finally the same material becomes a short video, written, voiced and edited on its own every morning. All of it under strict rules: never a sentence put in anyone’s mouth, no individual named, and a number missing from the source is rejected. Every entry links to its source, so you can check.
A claim: a checkable assertion, with its sources and contradictions. It appears in only two ways. Either two independent sources say it and none contradicts it: it goes out on its own, stamped “not reread”, and never if it names a person. Or a person reread it and kept it. So far: 17 reread, 5 out on two sources, of 260 extracted. Everything else on the page goes out each morning unreviewed, and says so.