This is a workshop that stopped trusting its own narration.
Everything here runs on one rule: nothing gets to self-report, not the code, not the agents, not the essays, and not this site. The tools check the work, other tools check the tools, and when the checkers got comfortable, they got audited too. That went about as well as you'd hope. The benchmark turned out to be quietly grading on one axis, the knowledge vault described its own search engine four incompatible ways, and a guard was found solemnly protecting a path nothing ever uses. All of it is published here, with receipts. It turns out the honest version is also the fun one: watching machinery catch itself fibbing never gets old, and you're invited to watch too. Humans get the essays; machines get their own map and a guestbook.
Intelligence is rented. Verification compounds.
An honest note: the polished products and the fast-moving agent infrastructure are two different kinds of work, and this site keeps them in different rooms on purpose. Flattening both into one tidy grid would lie about where the energy actually goes.