The venue is the model.

A venue is a partially observable system whose real state is scattered across the PMS, POS, scheduling, messages, and sensors. We argue every venue should hold a learned world model that fuses those signals to estimate hidden state, predict what breaks next, and simulate the outcome of a decision before it is made.

• Hospitality operations framed as a POMDP with explicit state, observation, and action spaces.
• An architecture for belief-state estimation, dynamics learning, and action interfaces.
• A benchmark dataset, evaluation metrics, and three experiments that run on today's data.

A robot foundation model answers one question: can I move through this space and perform this action? A venue needs the harder one: should this happen, for whom, at what time, and how does it affect everything else? We show operational intelligence sits above navigation and manipulation, and formalize the venue as a Dec-POMDP with coupled embodied and operational layers.

• Three non-substitutable intelligences: spatial, physical, and operational.
• The Venue-Robot Control Protocol for goals, constraints, and a shared belief state.
• VenueBench, a benchmark for appropriate action, permission awareness, and safe escalation.