What's True Now? ONTOVELA's World Model Separates Observed, Simulated, and Inferred States

Many digital twin platforms begin with visualization: building models, connecting devices, displaying metrics, and presenting reality through 3D interfaces. However, for autonomous agents, robots, and enterprise systems that must make decisions, the real difficulty is not displaying an object but determining what is actually true about that object right now.

Enterprise operational state is scattered across IoT platforms, CMDBs, business databases, process systems, robot systems, and agent private contexts. The same object may have multiple simultaneous values, each with different timestamps, sources, and trust levels. Keeping only a single "latest value" obscures critical distinctions:

Values directly observed by sensors

Values reported by personnel or external systems

Values computed from other data via deterministic rules

Values inferred by models or reasoning processes

Predictions about future states

Values that exist only inside simulation experiments

ONTOVELA defines itself as a Digital Enterprise Twin and Operational World Model Platform. It organizes physical, digital, organizational, process, agent, and robot states into an evidenced temporal graph, providing a unified reality foundation for planning, simulation, intervention, and closed-loop autonomy.

Strict State Classification

ONTOVELA categorizes state into six mutually exclusive types:

Observed : directly observed by devices or systems

Reported : reported by human operators or external actors

Derived : computed according to deterministic rules

Inferred : produced by models or inference processes

Predicted : forecasts about future states

Simulated : states that exist only in a simulated world

These categories cannot impersonate each other. In particular, a Simulated State must never be resolved directly as a real operational state, otherwise an agent could mistake a hypothetical experimental result for a real-world fact.

Bitemporal Model

ONTOVELA employs a dual-time model:

Event Time : when the event actually occurred in reality

System Time : when the platform learned and stored the information

This enables the system to answer three distinct questions:

What happened in reality at a given moment?

What did the system know given the information available at that time?

Why did later corrections or backfilled data change the current understanding?

When multiple sources provide conflicting states, the platform does not simply overwrite with the last write. Instead, it resolves based on source authority, freshness, and intended use, and can explicitly return Unknown or Conflicted.

Reality Views and Signed Snapshots

For concrete tasks, ONTOVELA provides Reality Views : limited, purpose-bound slices of reality. A planner does not need to read the entire enterprise world; it declares which objects, attributes, freshness requirements, and evidence conditions it needs.

The system can also generate signed snapshots . A plan or decision references not a constantly changing database query but a verifiable, comparable, auditable version of reality.

Product Division

This architecture yields a clear product separation:

GNOSIVELA expresses relatively stable enterprise knowledge

MNEMOVELA stores agent memory and cognitive state

PEIRAVELA creates controlled possible worlds

ONTOVELA maintains operational reality with provenance and temporal semantics

The world model is not a single mirror of reality, but a verifiable distinction among observation, report, inference, prediction, and simulation.

When agents begin acting on enterprise state, "data is latest" is no longer sufficient. The system must also know where the data came from, when it held true, whether it conflicts, whether it is still fresh, and whether it belongs to reality or simulation.

Project repository: https://github.com/axisrobo/ontovela-open