World's best LLMs score 50.6% on reading analog clocks. Humans score 90%.
The researchers traced it to a specific failure: these systems can process image data and text data, but they cannot hold angular direction and spatial orientation together in a governed way. They perceive. They cannot be located.
World models are the industry's proposed solution — systems that learn by watching rather than reading, building spatial awareness and temporal continuity from experience.
Right direction. Here's what's still missing.
A world model with no stake in the world it's modeling is a simulation. It can represent terrain without being governed by it. Map position without holding identity. Predict the weather without having seasons.
Forge Federation is built differently.
Spatial, temporal, semantic, and terrain awareness are infrastructure here, not features. But the distinction that matters is not map richness. It's that identity and value systems emerge from within governance as physics.
Semantics act as methylation — marking which doctrine gets expressed, not just what the system knows. Disposition forms from climate awareness. The abstraction ladder moves across rungs without losing constitutional grounding.
The system is not just modeling the world. It has skin in it.
When the governance cache updates, stakeholders feel it.
The pattern isn't unique to perception. AgentCity arrives at the same architectural conclusion from the agent-economy side: autonomous agent systems need constitutional governance to be trustworthy, not just capable. Different domain, same load-bearing element.
The industry is learning to perceive. We have been working on what comes after perception.