Why we built PolyBridge as a world model.

The questions you have aren't the questions markets ask

If you're an analyst or a strategist, the questions on your mind rarely match the markets that exist. You want to know whether oil rallies 10% in the next two months. The market asks whether oil hits $130 by July. You want to know whether a ceasefire holds through the year. The market asks whether a specific named leader is in office on a specific date. The questions are related, often closely, but they aren't the same.

The gap between questions decision-makers have and questions markets price is usually bridged by analysts doing the translation in their heads. They look at a basket of related markets, weigh them by relevance, factor in some context, and produce a judgment. The judgment isn't auditable or reproducible, and it doesn't scale beyond an analyst's working memory. We built PolyBridge to do the same translation explicitly.

What we mean by "world model"

A note on the term. "World model" is currently used in AI for two substantially different things. One is generative neural networks (DreamerV3, Sora, V-JEPA) trained to predict observations, systems that learn the world's structure by watching it unfold. The other is the older sense we use here: an explicit representation of what's going on, with named concepts and causal relationships between them. PolyBridge is the second kind. We've written separately about the technical distinction for readers who want it.

The PolyBridge world model is something like the mental model a senior macro analyst carries in their head. Inflation regime affects Fed stance. Fed stance affects risk sentiment. Geopolitical escalation affects oil supply. These relationships aren't speculative; they're well-understood, and analysts use them every day to reason about questions markets don't price directly.

The difference is that PolyBridge writes them down. It maintains a small map of how the world works: a few dozen named concepts and the causal connections between them. Every prediction market, every economic release, every news event becomes a noisy measurement of those underlying concepts. The model continuously infers the state of the world from those measurements, and any question you ask is answered by routing through the structure.

Five things this gets you

What this looks like in practice

And you can audit the chain. Every number traces back to capital-backed market signals and other observable data, which means risk committees, technical reviewers, and customer engineering teams can check the work.

The deeper bet

We built PolyBridge as a world model because we think the right way to do probabilistic intelligence for high-stakes decisions is to make the structure of the world explicit, rather than learning it implicitly inside a neural network whose outputs aren't auditable. Prediction markets are a good signal to drive this: there's capital on the line, so the prices reflect something other than commentary, and most other data sources have weaker truth-tracking properties. The architecture combines a structured world model with capital-backed signal and continuous updating, which is what institutional and defense customers ask for when they describe what they need: a system whose answers can be defended in front of a committee.

The technical foundation is decades of probabilistic AI research in the Pearl, Koller, and Friedman tradition of causal graphical models, updated with modern tools that solve the problems that limited earlier systems. The product is shaped by what institutional decision-makers ask for when they're being honest about what they need.

If you're making consequential decisions under uncertainty and your current tools give you point estimates without uncertainty, judgments without traceability, or answers only to questions someone else has framed, we'd be glad to talk.