World models and digital twins¶
An agent that reasons about an enterprise needs more than retrieved text. It needs a model of what exists, what is true, what changed, and what evidence supports the claim.
Operational world model¶
An operational world model is a shared representation of the state relevant to decisions: assets, orders, facilities, missions, people, policies, and their relationships. It is not a dashboard and not a data lake. It is a model that can answer questions.
Digital twin¶
A digital twin is a world model for a specific system — a factory, a fleet, a supply chain — kept in sync with the real system through assertions and observations. The twin is useful when it is temporal and evidence-bearing:
- Temporal (bitemporal): it distinguishes when something was true from when the system recorded it.
- Evidence-bearing: every assertion points to its source and confidence.
Project: ONTOVELA.
Concepts and grounding¶
A world model needs agreed definitions. Without them, “customer,” “site,” or “active” mean different things in different systems. A semantic layer defines concepts, resolves entities, evidence-backs claims, and governs conflicts.
Project: GNOSIVELA.
Possible worlds¶
A world model describes what is. A simulation explores what could be. A governed experiment defines worlds, scenarios, perturbations, controls, and seeds, then produces evidence that can be compared and verified.
Project: PEIRAVELA.
Why it matters for agents¶
- Shared context. Agents, planners, and humans reason over the same state.
- Grounded decisions. Claims carry evidence instead of confident prose.
- Safe experimentation. High-risk changes can be tested in a possible world before they touch the real one.
- Auditability. The history of beliefs is part of the record.
A caution¶
A world model is only as good as its ingestion and governance. A twin that silently accepts every assertion is a very expensive dashboard. Evidence, conflict handling, and freshness are not optional.