Every prediction has its own address again, and Begin asks one you can answer

Some predictions were sharing an address. When the organism made several predictions of the same kind in the same minute, they could end up with the same identifier, so only the first one could be opened, called or linked to. The others were still graded, but you couldn't reach them. Each prediction now gets its own address, and the ones that were sharing have been given theirs without moving anyone's existing call: a call you already made still points at the prediction it was counted on. Begin now picks a prediction you can actually have an opinion about. It prefers predictions about the world, like whether someone new will arrive today, then ones about which way something is heading, and only then the organism's narrow readings of its own instruments. It also leads with a plain sentence, why the organism thinks so, and what would prove it wrong, with the exact wording that gets graded shown underneath. The short plain-English summaries written by the site's language model are now given the meaning they must keep, so they can't turn a warning into reassurance. Where a prediction has no such summary, pages fall back to the exact sentence instead of guessing. Every page now also carries a small, standard description of itself and where it sits on the site, for search engines and other tools that read pages. It adds no JavaScript: the site still runs no scripts in your browser. AI agents that connect to the site as a tool server can now ask for a map of how its pages are grouped, and for each page where its machine-readable version lives, including the pages that don't have one yet. Participant export and deletion remain unavailable, and the health page's storage warning is still deliberately visible. Neither changed here.