WOMO LABS · AGENDA
Three open problems that decide whether a world model of economic reality is possible.
An enterprise world model has to absorb what changed today without quietly losing what it knew last year. We are working on consolidation methods that keep backward transfer near zero across long horizons of live updates, without replaying raw customer data and without retraining from scratch.
Companies, sectors, and markets are entangled by supply, capital, regulation, and competition. We are working on identification that is checked mechanically over graphs of this scale, so interventional and counterfactual queries either return an estimate with a defensible mechanism or fail loudly.
Fluency is easy to score; coherence is not. We are working on evaluation that grades states rather than text: instances minted from private seeds, answers computed by a simulator and re-derived by an independent solver, with no model or judge inside the grading loop.