World Models
ArcFlow turns physical spaces into queryable World Models
A World Model is a continuous, structured representation of everything happening in a physical space: entities, positions, relationships, events, and confidence scores. It updates at frame rate. It runs on every platform simultaneously — edge GPU, browser WASM, mobile, cloud. It is what every OZ product is built on.

The World Model is not video. It is a structured ArcFlow graph: every entity in the space (players, objects, zones, cameras) as a node, every relationship as an edge, every event as a timestamped transition. Each node carries position, velocity, confidence, and provenance — which sensor produced it, which model version processed it, and how certain the system is.
The graph is queryable in WorldCypher from any device. Live views update automatically as the graph changes. Operators see what the AI sees, in real time, with confidence scores attached to every claim.
Every tracked object as a graph node: 3D position, velocity, acceleration, orientation, confidence scores, and temporal history.
Spatial regions with semantic meaning — penalty areas, coverage footprints, restricted zones.
Timestamped transitions: entity entries and exits, state changes, anomalies, domain-specific events.
Every claim has a confidence score and source attribution.
Every data point traces back to sensor source, model version, and calibration ID.
WorldCypher live views auto-maintain as the graph changes.
Data sources
Edge capture → ArcFlow graph → any device
World Models are populated by ArcFlow sensors deployed at the venue edge. OZ Astro provides wide-scene context and entity detection from a single mounting point. OZ VI Venue adds the Precision Capture Plane — six robotic cameras with optical zoom, fusing coarse tracks into identification-grade spatial data. The graph syncs to ArcFlow Cloud and is accessible as scoped fragments on any device.
Operator view
Entity browser, event stream, validation checklist
OZ World is the operator interface to the live World Model. An entity browser shows every tracked entity with position, confidence, and last-seen timestamp. An event stream shows what the AI has detected in the past 24 hours. A validation checklist provides a pre-match gate: all entities calibrated, all cameras at threshold, no stale entries. One number — match-ready or not.

Your venue as a live, queryable graph of entities, positions, relationships, and events. Powered by ArcFlow. Accessible from any device. The intelligence layer that makes everything else...