A multi-part build: from why a few good rules beat a mountain of data, to a playable lending world that judges your AI.
Why we build practice worlds for regulated AI, learned from games. The story behind the series, and what it explores: costs, tradeoffs, and worlds grown from the law itself.
Regulated industries break world models in four specific ways: colliding clocks, feedback loops, legal fences, and your own output becoming your next input. They also hand you the one thing world-model builders everywhere else lack, which is a supply of free, dated, published interventions to test against.
Banking ran our definition-change test for us, better documented than we could have managed, and the answer was not the one anybody expects. On one day in January 2020 a single accounting rule moved reserves up at three banks and down at a fourth, and up and down inside the same bank. That gives a coherence test costing one group-by, and it also breaks a category we had been treating as one thing.
A handful of procedural rules can grow a whole living world. First a planet from a seed, then life that adapts to it, procedurally, and with an LLM. This is how games can teach our agents.
We let an LLM design competing species, run them headless in the living world, and refine them from the results. Over three rounds it revived an extinct lineage into the winner and abandoned its own planner. Going deeper on Part 1.
Interactive: build the lending game one mechanic at a time, then play the 30-year sim.